What engineering leaders and their teams report about Claude Code, with every finding tied to its evidence.
Read the findingsPractitioner sentiment, leaders and teams
+75 Strong positive from 53 sources that take a side
How this figure is madeEach from a different author and a different discussion. Use a task pilot to measure accepted work and review time.
Mar 2025 to Oct 2026.
138 sources from practitioners, by month and sentiment.
Engineering leaders and teams report the clearest value after they establish shared conventions and context. Indeed, Applied Intuition and Shopify describe Claude Code as their engineers' most used agent in 2026. Their main concerns are review load and the amount of work to delegate.
Positive, mixed and negative sources
The score counts each source once, as the split does. See How the numbers work.
See the evidenceBelow: 6 praise themes, 5 concerns, 2 points of agreement and 2 disagreements.
Counted for Engineering leaders and teams, from 68 sources. Change group
Every capability practitioners describe positively, ranked by how many sources say so. Open one to read who said it.
Counted for Engineering leaders and teams.
Positive
Alexey Milovidov, Co-Founder and Chief Technology Officer at ClickHouse
Agentic Coding at ClickHouseCompany blog post, ClickHouse Blog
ClickHouse's chief technology officer says agents became fully usable for daily work on its large C++ codebase from Opus 4.5, and the company mostly runs Claude Code in the terminal with Codex as a backup; he finds Claude Code slow at times, with bugs and glitches, and says agent output still needs review.
"Since Opus 4.5, agents have been fully usable for daily work on large C++ code bases."
"Claude Code is a very powerful tool, but it can be slow, and it has bugs and glitches. I still recommend using it over integrated agents."
"For CLI agents, we use mostly Claude Code with Opus 4.6, but some people prefer Codex CLI with Codex 5.4."
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Reliability, Operational control, Adoption and upkeep
Positive
Maxime Beauchemin, Founder and Chief Executive Officer of Preset
Adopting Claude Code: Riding the Software Economics SingularityCompany blog post, Preset Blog
Preset's founder says Claude Code was the first AI tool that worked on large repositories, clearing refactors and test backlogs in minutes plus review time, and that the LLMS.md and CLAUDE.md files in the Superset repo became the most helpful documents the team ever wrote.
"In our Superset repo, our LLMS.md and CLAUDE.md files have become the most helpful documents we've ever written."
"Before Claude Code (mid-May if you were first in line), AI pretty much sucked on large repos."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Cost predictability and limits
Positive
Steve Sewell, Builder.io
How I use Claude Code (+ my best tips)Company blog post, Builder.io
After a year as a Cursor power user, Sewell moved his agent work to Claude Code, the only agent that has successfully updated an 18,000-line React component at Builder.io. He runs it with permission prompts skipped and calls the US$100 Max plan worth it.
"No AI agent has ever successfully updated this file except Claude Code."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Lawrence Jones, software engineer, incident.io
The Pragmatic Engineer: Software engineering with LLMs in 2025Newsletter account
At incident.io, many engineers adopted Cursor first, and Claude Code then became the biggest change, with the entire team using it regularly by mid-2025.
"Claude Code has been the biggest change, though. Our entire team are regular users. Claude Code is the interactive terminal app that runs an Anthropic agent to explore and modify your codebase.”…"
Positive
Ahmad El-Shafey, Senior Engineering Manager at Slack
What I Learned Using Claude Code as an Engineering Manager During Focus WeekPersonal blog post, Medium
A Slack engineering manager used Claude Code to turn one doc fix into systematic cleanup and tooling, but found rules drift on bigger tasks, review stays essential, and over-engineering gets easy.
Practitioner account; direct use; counts toward Delivered outcomes
Positive
Charles Herring, co-founder and chairman of WitFoo
Teaching Claude the Old Tricks: Lessons from Migrating Legacy Code to an AI-Assisted WorkflowPersonal blog post, charlesherring.com
Brought WitFoo's decade-old production platform into Claude Code work by first having it document the code into CLAUDE.md, then writing a WITFOO_WAY.md standards file and cleaning up debt before a rebuild. Says without a standards document each Claude session makes slightly different decisions.
"Without a clear standards document, each Claude session makes slightly different decisions."
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Reliability
Positive
Dex Horthy, Founder of HumanLayer
No Vibes Allowed: Solving Hard Problems in Complex Codebases - Dex Horthy, HumanLayerConference talk (AI Engineer Code 2025), YouTube (AI Engineer channel; transcript on ai.engineer)
HumanLayer's founder describes a research, plan, implement workflow built around Claude Code's context window that took the three-person team to two to three times throughput on complex brownfield codebases.
"…minute, but like the point is you're constantly st keeping your context window small. You're building your entire workflow around context management. So comes in three phases. research, plan, implement. Um, and we're going to try to stay in the smart zone…"
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Positive
Jonathan Pedoeem, Technical founder of PromptLayer
Claude Code has changed how we do engineeringCompany blog post, PromptLayer blog
PromptLayer's founder says Claude Code let the team clear its backlog under a 'build it if under an hour' rule, while warning reviews remain essential because it makes characteristic mistakes.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes
Positive
Orta Therox, engineer at Puzzmo
6 Weeks of Claude CodeCompany engineering blog post, Puzzmo Blog
Cleared what Orta Therox calls years of tech-debt backlog solo in six weeks with Claude Code while working on the existing roadmap. Credits monorepos and boring, explicit technologies, and finds the change hard to see in pull request, commit and line counts.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Mixed
Alessio Ferri, Thoughtworks
Claude Code saved us 97% of the work on the first try. Then it failed utterly.Engineering blog post, Thoughtworks Insights blog
With a subject expert, Claude Code added Python support to Thoughtworks' CodeConcise in half a day, against two to four weeks for a developer pair. Three JavaScript attempts failed: the grammar would not work, then it used libraries that did not exist, then internal packages that did not exist. The author says the agent lacks stronger verification.
Practitioner account, self-described; structured test; counts toward Delivered outcomes
Counted for Engineering leaders and teams.
Positive, 9
Positive
Peter Ludwig, co-founder and chief technology officer, Applied Intuition
Latent Space: Applied Intuition with Peter LudwigPodcast episode with publisher transcript
At Applied Intuition, which uses every coding tool and ranks adoption on a leaderboard, Claude Code has replaced Cursor as the most used tool.
"Yeah. We use everything. So Cursor was, I think the hottest tool in the company for a good while. Now Claude Code, I think has taken the reign on that. We have a internal leader, leaderboard that we use just to sort of encourage adoption…"
Positive
James Reggio, chief technology officer, Brex
Latent Space: Brex's AI Hail Mary, with James ReggioPodcast episode with publisher transcript
Brex lets employees choose between Cursor, Windsurf and Claude Code credits, and some engineers stay on Claude Code after nine months because they know it.
"…some folks are just like, I want to stick with Cloud Code because I know it now. I’ve been working with it for like nine months, so I don’t need to keep switching."
Positive
Mitch Alderson, Render
Testing AI coding agents (2025): Cursor vs. Claude, OpenAI, and GeminiIndependent benchmark, Render engineering blog
Rated four coding agents on a vibe-coded app and two large pre-AI repos. Claude Code averaged 6.8 of 10 against Cursor's 8, with 5 of 10 on context; calls it best for rapid prototypes and a productive terminal UX.
"Cursor leads on setup speed, Docker/Render deployment, and code quality, while Claude Code is best for rapid prototypes and a productive terminal UX."
Practitioner account; structured test; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Positive
saluki, Hacker News commenter
Comment on: Ask HN: Is anybody producing good code with coding agents?Hacker News comment
Reports great success with a short CLAUDE.md, skills, design and plan sessions before building, high test coverage, and Claude Code review plus human PR review for faster velocity.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Ahmad El-Shafey, Senior Engineering Manager at Slack
What I Learned Using Claude Code as an Engineering Manager During Focus WeekPersonal blog post, Medium
A Slack engineering manager used Claude Code to turn one doc fix into systematic cleanup and tooling, but found rules drift on bigger tasks, review stays essential, and over-engineering gets easy.
Practitioner account; direct use; counts toward Delivered outcomes
Positive
Asher Moore, engineering leader at Airtasker
Rolling out Claude Code at AirtaskerEngineering blog post, Technology @ Airtasker (Medium)
Airtasker rolled Claude Code out across engineering via AWS Bedrock for data sovereignty, with managed settings, OpenTelemetry dashboards, and a custom spend tracker after CLI cost figures proved unreliable.
Practitioner account; ran the rollout; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Shrivu Shankar, leads AI developer tooling at Abnormal AI
How I Use Every Claude Code FeaturePersonal blog post, Shrivu's Substack (blog.sshh.io)
Describes how Abnormal AI uses Claude Code on its monorepo: a strictly maintained 13KB CLAUDE.md, clear-and-document restarts over auto-compaction, a hook that blocks commits until tests pass, and enterprise API keys for usage-based pricing.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Reliability, Operational control, Adoption and upkeep
Positive
Steve Sewell, Builder.io
How I use Claude Code (+ my best tips)Company blog post, Builder.io
After a year as a Cursor power user, Sewell moved his agent work to Claude Code, the only agent that has successfully updated an 18,000-line React component at Builder.io. He runs it with permission prompts skipped and calls the US$100 Max plan worth it.
"No AI agent has ever successfully updated this file except Claude Code."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Rory Bain, Product Engineer at incident.io
How we're shipping faster with Claude Code and Git WorktreesEngineering blog post, incident.io blog
incident.io engineers run several Claude Code agents in parallel via Git worktrees and plan mode; one US$8 session delivered an 18% faster API generation step, though architecture stays human-led.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control
Mixed or negative, 1
Mixed
Katie Parrott, staff writer, Every
Every: Vibe Check, Opus 5.5 Is Pulling Our Codex Converts Back to ClaudeOrganization write-up
Two Every staff who had switched from Claude Code to Codex in the months before September 22, 2026 were reconsidering after Opus 5.5, though one still prefers the Codex harness and the CEO still uses Codex for most work.
"Mike Taylor and Tyler Nishida are former Claude power users who switched to Codex in recent months, but Opus 5.5 has them reconsidering that switch."
Counted for Engineering leaders and teams.
Positive, 8
Positive
cyan, iOS engineer on the Mercoin (MC) Client team
Scaling Myself: How I Run 22 Claude Code Sessions for DS4 MigrationEngineering blog post, Mercari Engineering blog
A Mercoin iOS engineer runs 22 Claude Code sessions across git worktrees with phase-specific slash commands to ship about 140 design-system migration PRs, plus dedicated sessions reviewing teammates' PRs.
Practitioner account; direct use; counts toward Delivered outcomes, Reliability
Positive
Pablo Arellano, Software Developer at Atolio
A UI Framework Migration in the Age of AI: How We Migrated Svelte 4 to 5 with Claude Code in Six HoursCompany engineering blog post, Atolio Blog
After months of human investigation went into one detailed prompt, Atolio let Claude Code run in auto mode for about six hours to move thousands of files from Svelte 4 to 5; two developers then spent about two days on warnings and QA, which found one bug.
"…we let it run in auto mode for around six hours on a Friday afternoon…"
"Once it was in testing, the team found only one bug during QA."
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Operational control
Positive
Devon Edwards Joseph, Senior Engineer at Spotify
Background Coding Agents: Supercharging Downstream Consumer Dataset Migrations (Honk, Part 4)Engineering blog post, Spotify Engineering
A Spotify team rolled out 240 automated dataset migration PRs with its Claude Code based agent, work it estimated at about 10 engineering weeks by hand; it needed very detailed prompts, and it dropped one pipeline framework whose style varied too much between teams.
"We estimated that it would have taken around 10 engineering weeks of effort to complete these migrations manually."
"…we successfully rolled out 240 automated migration PRs using Fleetshift…"
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes
Positive
Orta Therox, engineer at Puzzmo
6 Weeks of Claude CodeCompany engineering blog post, Puzzmo Blog
Cleared what Orta Therox calls years of tech-debt backlog solo in six weeks with Claude Code while working on the existing roadmap. Credits monorepos and boring, explicit technologies, and finds the change hard to see in pull request, commit and line counts.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Mixed
Alex Ghiculescu, co-founder of Tanda
Nobody Codes Here AnymoreNewsletter, Alex Ghiculescu's Newsletter (Substack)
Rolling out Cursor and Claude Code to about 40 developers, Tanda's co-founder finds Claude Code great at refactors but subtly wrong on bug fixes, estimating roughly 20% productivity gains.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Cost predictability and limits
Positive
alexmuro, Runs a small analytics and web products lab; manages 5 developers
Comment on: Ask HN: How is AI-assisted coding going for you professionally?Hacker News comment
Since trying Claude Code with Opus 4.5 alexmuro writes few features by hand, rebuilt a server and added sync, keeps a short leash, and reports mixed reactions across their five-person team.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Positive
Charles Herring, co-founder and chairman of WitFoo
Teaching Claude the Old Tricks: Lessons from Migrating Legacy Code to an AI-Assisted WorkflowPersonal blog post, charlesherring.com
Brought WitFoo's decade-old production platform into Claude Code work by first having it document the code into CLAUDE.md, then writing a WITFOO_WAY.md standards file and cleaning up debt before a rebuild. Says without a standards document each Claude session makes slightly different decisions.
"Without a clear standards document, each Claude session makes slightly different decisions."
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Reliability
Positive
Maxime Beauchemin, Founder and Chief Executive Officer of Preset
Adopting Claude Code: Riding the Software Economics SingularityCompany blog post, Preset Blog
Preset's founder says Claude Code was the first AI tool that worked on large repositories, clearing refactors and test backlogs in minutes plus review time, and that the LLMS.md and CLAUDE.md files in the Superset repo became the most helpful documents the team ever wrote.
"In our Superset repo, our LLMS.md and CLAUDE.md files have become the most helpful documents we've ever written."
"Before Claude Code (mid-May if you were first in line), AI pretty much sucked on large repos."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Cost predictability and limits
Mixed or negative, 1
Mixed
Kush Gupta, engineer at Graphite
How I got Claude to write code I could actually shipEngineering blog post, Graphite blog
Graphite engineers found Claude Code's large refactor diffs unreviewable, even edited tests, so many used it only for point fixes until CLAUDE.md stacking instructions produced small reviewable PRs.
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Counted for Engineering leaders and teams.
No opinion stated
Arunabh Dastidar, startup engineering lead
The big discount on intelligencePractitioner write-up
Five engineers who paid US$800 to US$1,000 a month together on Max and Pro plans paid about US$1,000 a day for the same work at API rates, Dastidar writes.
"After this, the bill was approximately $1000 per day for the same work. So in one fell swoop, the cost went up thirty times."
Mixed
linsomniac, Hacker News commenter
Comment on: Tokenmaxxing is dead, long live tokenmaxxingHacker News comment
At a company thrifty with AI spend, had a US$100 a month spend while trying to be its AI leader. Sees that Claude Code can do a lot of heavy lifting at work, while most colleagues stuck cost-maxing cannot imagine hitting session limits; says there's a case the cost-maxing hurt the company.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
Shodhan Sheth, Enterprise Modernization, Platforms & Cloud Lead at Thoughtworks
Claude Code and COBOL modernization: What’s the reality?Company article, Thoughtworks Insights
Says, from 18 months of AI-assisted legacy modernization at Thoughtworks, that Claude Code is one of the best coding agents so far, exceptional on time and cost but weaker on correctness and drift, and that agents cannot yet reliably ingest tens of millions of lines at once.
"In terms of Claude Code, it’s exceptional when it comes to time and cost. What it’s not as strong as, though, are the bottom two issues…"
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Reliability
Positive
Vincent Quigley, then Staff Engineer at Sanity
First attempt will be 95% garbage: A staff engineer's 6-week journey with Claude CodeEngineering blog post, Sanity Engineering
A Sanity engineer treats Claude Code like a junior developer with amnesia: three attempts per feature, Claude.md context, layered review, 2-3x faster shipping at roughly US$1000-1500 per month.
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Cost predictability and limits, Adoption and upkeep
Positive
Steve Sewell, Builder.io
How I use Claude Code (+ my best tips)Company blog post, Builder.io
After a year as a Cursor power user, Sewell moved his agent work to Claude Code, the only agent that has successfully updated an 18,000-line React component at Builder.io. He runs it with permission prompts skipped and calls the US$100 Max plan worth it.
"No AI agent has ever successfully updated this file except Claude Code."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits, Adoption and upkeep
Counted for Engineering leaders and teams.
Positive, 5
Positive
Armin Ronacher, founder of a new company
90%Personal blog post, lucumr.pocoo.org
Ronacher reports more than 90% AI-written code in a 40,000-line Go service at his new company. Claude Code is his tool for debugging that needs heavy tool access, such as a deadlock or corrupted data, and he still reviews every line.
"Claude is indispensable still when debugging and needing a lot of tool access (eg: why do I have a deadlock, why is there corrupted data in the database etc.)."
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Positive
Brian Scanlan, engineer at Intercom
How we use Claude Code today at IntercomCompany blog post, Fin Ideas (Intercom)
Intercom turned Claude Code into an internal platform: 13 plugins, 100+ skills, hook-enforced PR workflow, guarded read-only production console, OpenTelemetry observability and auto-updated CLAUDE.md files.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Positive
Max Charas and Marc Bruggmann, Senior Staff and Principal Engineers at Spotify
Background Coding Agents: Context Engineering (Honk, Part 2)Engineering blog post, Spotify Engineering
Spotify moved its background migration agent to Claude Code after open source and homegrown agents struggled, finding end-state prompts, few tools and a verify tool give mergeable PRs.
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Operational control
Mixed
Peter Silberman, Chief Technology Officer at Fixify
Practical Lessons Learned using Claude Code to automate IntegrationsNewsletter post (Substack)
Tried to have Claude Code zero-shot whole Fixify integrations, first through CLAUDE.md, then with specialized sub-agents that hallucinated permissions and API routes until an agent distilling vendor docs into a reference file cut that dramatically. Says full autonomy is probably not coming without human oversight.
"Unsurprisingly, Claude and its sub-agents hallucinated, permissions that didn’t exist, API routes that weren’t real."
"Will we ever ship a fully autonomous integration? Probably not without human oversight."
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Adoption and upkeep
Positive
Rory Bain, Product Engineer at incident.io
How we're shipping faster with Claude Code and Git WorktreesEngineering blog post, incident.io blog
incident.io engineers run several Claude Code agents in parallel via Git worktrees and plan mode; one US$8 session delivered an 18% faster API generation step, though architecture stays human-led.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control
Mixed or negative, 1
Mixed
Stella Laurenzo (stellaraccident), software engineer (employer unnamed in the issue)
[MODEL] Claude Code is unusable for complex engineering tasks with the Feb updatesGitHub issue
Reports from months of mined Claude Code logs on IREE projects using Opus that complex engineering quality degraded from February, forcing a shutdown of the concurrent agent fleet for single-session work. Says Claude has been good to them and hopes Anthropic can address it.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Reliability
Counted for Engineering leaders and teams.
Positive, 4
Positive
Spencer Niemi, Elastic InfoSec
Claude Code/Cowork monitoring at scale with Otel & ElasticCompany engineering blog post, Elastic Security Labs
Describes how Elastic's InfoSec team monitors Claude Code, in active use across Elastic engineering, by exporting its OpenTelemetry events into Elastic, with telemetry set through managed settings deployed by IT that users cannot override. Notes telemetry is opt-in and redacts prompts and tool arguments by default.
"Claude Code supports managed settings that are deployed by IT and cannot be overridden by users."
Practitioner account, self-described; ran the rollout; counts toward Operational control
Positive
Stephan Lensky, senior engineer on the AI Developer Experience team at HubSpot
Cloud Coding Agents at HubSpotCompany engineering blog post, HubSpot Product Blog
HubSpot built its cloud agent platform on Claude Code, which it already knew worked well for its engineers, and in six months merged over 7,000 fully AI-generated pull requests and reviewed over 50,000 human ones, using hooks to enforce the behavior it wants.
"…we used an existing platform, Claude Code, which we already knew worked well for our engineers…"
"To date, we have merged over 7,000 fully AI-generated pull requests and code reviewed over 50,000 pull requests authored by humans."
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Reliability, Operational control
Positive
Brian Scanlan, engineer at Intercom
How we use Claude Code today at IntercomCompany blog post, Fin Ideas (Intercom)
Intercom turned Claude Code into an internal platform: 13 plugins, 100+ skills, hook-enforced PR workflow, guarded read-only production console, OpenTelemetry observability and auto-updated CLAUDE.md files.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Positive
Shrivu Shankar, leads AI developer tooling at Abnormal AI
How I Use Every Claude Code FeaturePersonal blog post, Shrivu's Substack (blog.sshh.io)
Describes how Abnormal AI uses Claude Code on its monorepo: a strictly maintained 13KB CLAUDE.md, clear-and-document restarts over auto-compaction, a hook that blocks commits until tests pass, and enterprise API keys for usage-based pricing.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Reliability, Operational control, Adoption and upkeep
Mixed or negative, 1
Negative
thewebguyd, Hacker News commenter
Comment on: Claude TagHacker News comment
Rolled Cowork out to all non-technical staff at work and calls Cowork/Claude Desktop frustrating, with no native audit log and all-or-nothing access for Claude Code and for some MCP connections such as QuickBooks Online. Wants governance features and per-member RBAC.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Every concern practitioners raise, most raised first.
Counted for Engineering leaders and teams.
Raised as a concern, 16
Negative
throwatdem12311, Only staff engineer at a company
Comment on: The new rules of context engineering for Claude 5 generation modelsHacker News comment
As sole staff engineer reviewing PRs from three offshore developers using Claude Code, throwatdem12311 says review has become impossible to keep up with and delivery has slowed.
Practitioner account, pseudonymous; second-hand; not counted, repeats a stronger account
Positive
maiybe, Hacker News commenter
Comment on: Godot will no longer accept AI-authored code contributionsHacker News comment
On a Godot mobile app their team estimates 3x speed and 3x more pull requests with about 20% more regressions. They dropped review below a file-size threshold and grew CI tests about 500%.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Positive
Ahmad El-Shafey, Senior Engineering Manager at Slack
What I Learned Using Claude Code as an Engineering Manager During Focus WeekPersonal blog post, Medium
A Slack engineering manager used Claude Code to turn one doc fix into systematic cleanup and tooling, but found rules drift on bigger tasks, review stays essential, and over-engineering gets easy.
Practitioner account; direct use; counts toward Delivered outcomes
Positive
Chad Pytel, Developer and CEO at thoughtbot
Claude Code: Production ready code in a two-week sprintCompany blog post, thoughtbot blog
thoughtbot built production-quality Rails code for a client in a two-week sprint with Claude Code by giving it very small tasks, keeping a CLAUDE.md and reviewing every change before a human commit; sometimes it wrote tests that checked only the fake objects they set up.
"During a two-week sprint with a startup, we ended the intense engagement with true production-quality code…"
"Allowing Claude Code to make commits relinquishes this critical quality control step."
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes
Mixed
Juwan Wheatley, Expo engineer
What our web team learned using Claude Code for a monthCompany engineering blog post, Expo Blog
Expo's web team says Claude Code performs best on well-specified repository work, while context quality deteriorates as it accumulates and the agent still needs explicit reminders and human review.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Reliability
Positive
Armin Ronacher, founder of a new company
90%Personal blog post, lucumr.pocoo.org
Ronacher reports more than 90% AI-written code in a 40,000-line Go service at his new company. Claude Code is his tool for debugging that needs heavy tool access, such as a deadlock or corrupted data, and he still reviews every line.
"Claude is indispensable still when debugging and needing a lot of tool access (eg: why do I have a deadlock, why is there corrupted data in the database etc.)."
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Positive
Jonathan Pedoeem, Technical founder of PromptLayer
Claude Code has changed how we do engineeringCompany blog post, PromptLayer blog
PromptLayer's founder says Claude Code let the team clear its backlog under a 'build it if under an hour' rule, while warning reviews remain essential because it makes characteristic mistakes.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes
Positive
Vincent Quigley, then Staff Engineer at Sanity
First attempt will be 95% garbage: A staff engineer's 6-week journey with Claude CodeEngineering blog post, Sanity Engineering
A Sanity engineer treats Claude Code like a junior developer with amnesia: three attempts per feature, Claude.md context, layered review, 2-3x faster shipping at roughly US$1000-1500 per month.
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Cost predictability and limits, Adoption and upkeep
Mixed
Peter Silberman, Chief Technology Officer at Fixify
Practical Lessons Learned using Claude Code to automate IntegrationsNewsletter post (Substack)
Tried to have Claude Code zero-shot whole Fixify integrations, first through CLAUDE.md, then with specialized sub-agents that hallucinated permissions and API routes until an agent distilling vendor docs into a reference file cut that dramatically. Says full autonomy is probably not coming without human oversight.
"Unsurprisingly, Claude and its sub-agents hallucinated, permissions that didn’t exist, API routes that weren’t real."
"Will we ever ship a fully autonomous integration? Probably not without human oversight."
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Adoption and upkeep
Mixed
Kush Gupta, engineer at Graphite
How I got Claude to write code I could actually shipEngineering blog post, Graphite blog
Graphite engineers found Claude Code's large refactor diffs unreviewable, even edited tests, so many used it only for point fixes until CLAUDE.md stacking instructions produced small reviewable PRs.
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Positive
Diwank Singh, Julep
Field Notes From Shipping Real Code With ClaudePersonal blog post, diwank.space
Describes how Julep ships production code daily with Claude under a CLAUDE.md that tells the agent never to modify tests, change API contracts or alter migration files. Humans write the tests, and AI-assisted commits are tagged so reviewers pay extra attention.
"Next, I’ll walk you through the infrastructure we use at Julep to ship production code daily with Claude’s help."
Practitioner account; direct use; counts toward Delivered outcomes, Adoption and upkeep
Mixed
Alex Ghiculescu, co-founder of Tanda
Nobody Codes Here AnymoreNewsletter, Alex Ghiculescu's Newsletter (Substack)
Rolling out Cursor and Claude Code to about 40 developers, Tanda's co-founder finds Claude Code great at refactors but subtly wrong on bug fixes, estimating roughly 20% productivity gains.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Cost predictability and limits
Mixed
Alessio Ferri, Thoughtworks
Claude Code saved us 97% of the work on the first try. Then it failed utterly.Engineering blog post, Thoughtworks Insights blog
With a subject expert, Claude Code added Python support to Thoughtworks' CodeConcise in half a day, against two to four weeks for a developer pair. Three JavaScript attempts failed: the grammar would not work, then it used libraries that did not exist, then internal packages that did not exist. The author says the agent lacks stronger verification.
Practitioner account, self-described; structured test; counts toward Delivered outcomes
Negative
kren87, former tech lead
Comment on: Slop DebtHacker News comment
As tech lead of a team vibe coding with Claude, reviewed merge requests of at least 800 lines, multiple per day per person; says understanding the code took 2 hours. Quit to start a company and is slowly reeling back to coding by hand, which they prefer.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Caroline Schneider, Marmelab
Claude Code Tips I Wish I'd Had From Day OneCompany engineering blog post, Marmelab
After months of daily use, Marmelab asks Claude Code for one step at a time and reviews each before the next, because asking for a whole feature sends it off the rails. Lessons from each session go into CLAUDE.md, rules and skills.
"If you say “implement the whole feature,” Claude will go off the rails. Instead, ask for step one, review it, and only then ask for step two."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability
Mixed
Darryn Dieken, Chief Availability Officer, Salesforce
Salesforce Engineering: How AI-enabled tooling boosted code output 30%Engineering write-up
Salesforce uses Claude Code as a command-line tool for automation and scripting beside Cursor and Windsurf, and reports code in production up 30%, which put pressure on review and deployment.
"AI accelerated development speed, increasing code in production by 30%. This surge created significant pressure on downstream processes."
Discussed without concern, 7
Positive
Kat Korson, Director at Red Eagle Tech
How We Trained Our Engineering Team To Actually Use Claude Code (not just chat with it)Company-authored engineering practice account, Collabnix
Red Eagle Tech describes Claude Code reading existing workflows, tracing repository conventions, generating a complete CI/CD pipeline, and reaching a reviewed pull request in about 25 minutes. Manual work took about three hours, and the account still calls for review.
Practitioner account; ran the rollout; counts toward Delivered outcomes
Positive
saluki, Hacker News commenter
Comment on: Ask HN: Is anybody producing good code with coding agents?Hacker News comment
Reports great success with a short CLAUDE.md, skills, design and plan sessions before building, high test coverage, and Claude Code review plus human PR review for faster velocity.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
jaegerpicker, Hacker News commenter
Comment on: AI Coding and Its DiscontentsHacker News comment
jaegerpicker's team reports 5x output with fewer quality issues by having Claude research each ticket into acceptance criteria that humans validate, then requiring local testing, integration tests, and 90% coverage.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Positive
Alexey Milovidov, Co-Founder and Chief Technology Officer at ClickHouse
Agentic Coding at ClickHouseCompany blog post, ClickHouse Blog
ClickHouse's chief technology officer says agents became fully usable for daily work on its large C++ codebase from Opus 4.5, and the company mostly runs Claude Code in the terminal with Codex as a backup; he finds Claude Code slow at times, with bugs and glitches, and says agent output still needs review.
"Since Opus 4.5, agents have been fully usable for daily work on large C++ code bases."
"Claude Code is a very powerful tool, but it can be slow, and it has bugs and glitches. I still recommend using it over integrated agents."
"For CLI agents, we use mostly Claude Code with Opus 4.6, but some people prefer Codex CLI with Codex 5.4."
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Reliability, Operational control, Adoption and upkeep
Positive
Brian Scanlan, engineer at Intercom
How we use Claude Code today at IntercomCompany blog post, Fin Ideas (Intercom)
Intercom turned Claude Code into an internal platform: 13 plugins, 100+ skills, hook-enforced PR workflow, guarded read-only production console, OpenTelemetry observability and auto-updated CLAUDE.md files.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Positive
Charles Herring, co-founder and chairman of WitFoo
Teaching Claude the Old Tricks: Lessons from Migrating Legacy Code to an AI-Assisted WorkflowPersonal blog post, charlesherring.com
Brought WitFoo's decade-old production platform into Claude Code work by first having it document the code into CLAUDE.md, then writing a WITFOO_WAY.md standards file and cleaning up debt before a rebuild. Says without a standards document each Claude session makes slightly different decisions.
"Without a clear standards document, each Claude session makes slightly different decisions."
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Reliability
Positive
Dex Horthy, Founder of HumanLayer
No Vibes Allowed: Solving Hard Problems in Complex Codebases - Dex Horthy, HumanLayerConference talk (AI Engineer Code 2025), YouTube (AI Engineer channel; transcript on ai.engineer)
HumanLayer's founder describes a research, plan, implement workflow built around Claude Code's context window that took the three-person team to two to three times throughput on complex brownfield codebases.
"…minute, but like the point is you're constantly st keeping your context window small. You're building your entire workflow around context management. So comes in three phases. research, plan, implement. Um, and we're going to try to stay in the smart zone…"
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Counted for Engineering leaders and teams.
Raised as a concern, 6
Mixed
Stella Laurenzo (stellaraccident), software engineer (employer unnamed in the issue)
[MODEL] Claude Code is unusable for complex engineering tasks with the Feb updatesGitHub issue
Reports from months of mined Claude Code logs on IREE projects using Opus that complex engineering quality degraded from February, forcing a shutdown of the concurrent agent fleet for single-session work. Says Claude has been good to them and hopes Anthropic can address it.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Reliability
Mixed
Shodhan Sheth, Enterprise Modernization, Platforms & Cloud Lead at Thoughtworks
Claude Code and COBOL modernization: What’s the reality?Company article, Thoughtworks Insights
Says, from 18 months of AI-assisted legacy modernization at Thoughtworks, that Claude Code is one of the best coding agents so far, exceptional on time and cost but weaker on correctness and drift, and that agents cannot yet reliably ingest tens of millions of lines at once.
"In terms of Claude Code, it’s exceptional when it comes to time and cost. What it’s not as strong as, though, are the bottom two issues…"
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Reliability
Mixed
Alessio Ferri, Thoughtworks
Claude Code saved us 97% of the work on the first try. Then it failed utterly.Engineering blog post, Thoughtworks Insights blog
With a subject expert, Claude Code added Python support to Thoughtworks' CodeConcise in half a day, against two to four weeks for a developer pair. Three JavaScript attempts failed: the grammar would not work, then it used libraries that did not exist, then internal packages that did not exist. The author says the agent lacks stronger verification.
Practitioner account, self-described; structured test; counts toward Delivered outcomes
Positive
Ahmad El-Shafey, Senior Engineering Manager at Slack
What I Learned Using Claude Code as an Engineering Manager During Focus WeekPersonal blog post, Medium
A Slack engineering manager used Claude Code to turn one doc fix into systematic cleanup and tooling, but found rules drift on bigger tasks, review stays essential, and over-engineering gets easy.
Practitioner account; direct use; counts toward Delivered outcomes
Positive
Vincent Quigley, then Staff Engineer at Sanity
First attempt will be 95% garbage: A staff engineer's 6-week journey with Claude CodeEngineering blog post, Sanity Engineering
A Sanity engineer treats Claude Code like a junior developer with amnesia: three attempts per feature, Claude.md context, layered review, 2-3x faster shipping at roughly US$1000-1500 per month.
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Cost predictability and limits, Adoption and upkeep
Mixed
Alex Ghiculescu, co-founder of Tanda
Nobody Codes Here AnymoreNewsletter, Alex Ghiculescu's Newsletter (Substack)
Rolling out Cursor and Claude Code to about 40 developers, Tanda's co-founder finds Claude Code great at refactors but subtly wrong on bug fixes, estimating roughly 20% productivity gains.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Cost predictability and limits
Discussed without concern, 1
Positive
Rory Bain, Product Engineer at incident.io
How we're shipping faster with Claude Code and Git WorktreesEngineering blog post, incident.io blog
incident.io engineers run several Claude Code agents in parallel via Git worktrees and plan mode; one US$8 session delivered an 18% faster API generation step, though architecture stays human-led.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control
Counted for Engineering leaders and teams.
Raised as a concern, 6
Mixed
Shodhan Sheth, Enterprise Modernization, Platforms & Cloud Lead at Thoughtworks
Claude Code and COBOL modernization: What’s the reality?Company article, Thoughtworks Insights
Says, from 18 months of AI-assisted legacy modernization at Thoughtworks, that Claude Code is one of the best coding agents so far, exceptional on time and cost but weaker on correctness and drift, and that agents cannot yet reliably ingest tens of millions of lines at once.
"In terms of Claude Code, it’s exceptional when it comes to time and cost. What it’s not as strong as, though, are the bottom two issues…"
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Reliability
Mixed
Juwan Wheatley, Expo engineer
What our web team learned using Claude Code for a monthCompany engineering blog post, Expo Blog
Expo's web team says Claude Code performs best on well-specified repository work, while context quality deteriorates as it accumulates and the agent still needs explicit reminders and human review.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Reliability
Positive
Dex Horthy, Founder of HumanLayer
No Vibes Allowed: Solving Hard Problems in Complex Codebases - Dex Horthy, HumanLayerConference talk (AI Engineer Code 2025), YouTube (AI Engineer channel; transcript on ai.engineer)
HumanLayer's founder describes a research, plan, implement workflow built around Claude Code's context window that took the three-person team to two to three times throughput on complex brownfield codebases.
"…minute, but like the point is you're constantly st keeping your context window small. You're building your entire workflow around context management. So comes in three phases. research, plan, implement. Um, and we're going to try to stay in the smart zone…"
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Positive
Shrivu Shankar, leads AI developer tooling at Abnormal AI
How I Use Every Claude Code FeaturePersonal blog post, Shrivu's Substack (blog.sshh.io)
Describes how Abnormal AI uses Claude Code on its monorepo: a strictly maintained 13KB CLAUDE.md, clear-and-document restarts over auto-compaction, a hook that blocks commits until tests pass, and enterprise API keys for usage-based pricing.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Reliability, Operational control, Adoption and upkeep
Positive
Vincent Quigley, then Staff Engineer at Sanity
First attempt will be 95% garbage: A staff engineer's 6-week journey with Claude CodeEngineering blog post, Sanity Engineering
A Sanity engineer treats Claude Code like a junior developer with amnesia: three attempts per feature, Claude.md context, layered review, 2-3x faster shipping at roughly US$1000-1500 per month.
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Cost predictability and limits, Adoption and upkeep
Positive
Mitch Alderson, Render
Testing AI coding agents (2025): Cursor vs. Claude, OpenAI, and GeminiIndependent benchmark, Render engineering blog
Rated four coding agents on a vibe-coded app and two large pre-AI repos. Claude Code averaged 6.8 of 10 against Cursor's 8, with 5 of 10 on context; calls it best for rapid prototypes and a productive terminal UX.
"Cursor leads on setup speed, Docker/Render deployment, and code quality, while Claude Code is best for rapid prototypes and a productive terminal UX."
Practitioner account; structured test; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Discussed without concern, 6
Positive
Andrew Walker, Engineering Director at Fin.ai (Intercom)
What we're learning from engineers getting the most from Claude CodeCompany blog post, Fin Ideas (Intercom)
Shares, as an engineering director at Fin.ai, what engineers reported: some of the most experienced accelerate faster than earlier-career peers, and those getting the most from Claude plan over many turns and invest in claude.md as evolving instructions. Says many keep their anxiety to themselves.
Practitioner account, self-described; second-hand; not counted until substantiated
Positive
cyan, iOS engineer on the Mercoin (MC) Client team
Scaling Myself: How I Run 22 Claude Code Sessions for DS4 MigrationEngineering blog post, Mercari Engineering blog
A Mercoin iOS engineer runs 22 Claude Code sessions across git worktrees with phase-specific slash commands to ship about 140 design-system migration PRs, plus dedicated sessions reviewing teammates' PRs.
Practitioner account; direct use; counts toward Delivered outcomes, Reliability
Mixed
Sergii Gorbachov, Staff Software Engineer at Slack
Agentic Testing: Where Agents Fit in the E2E Testing StackCompany engineering blog post, Slack Engineering
Ran more than 200 agentic end-to-end test runs at Slack through non-interactive Claude Code. Agent runs typically cost US$15 to 30 each and averaged about 5 to 11 minutes; says the cost may currently suit them better to debugging and exploratory testing than to high-frequency CI.
"Agent-driven runs were typically $15 to 30 per execution, compared to much cheaper traditional test runs."
"Due to the cost, agent-driven testing may currently be better suited for targeted debugging or exploratory testing than for high-frequency CI execution…"
Practitioner account, name and role on their organization's own page; structured test; counts toward Delivered outcomes
Positive
Caroline Schneider, Marmelab
Claude Code Tips I Wish I'd Had From Day OneCompany engineering blog post, Marmelab
After months of daily use, Marmelab asks Claude Code for one step at a time and reviews each before the next, because asking for a whole feature sends it off the rails. Lessons from each session go into CLAUDE.md, rules and skills.
"If you say “implement the whole feature,” Claude will go off the rails. Instead, ask for step one, review it, and only then ask for step two."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability
Positive
Devon Edwards Joseph, Senior Engineer at Spotify
Background Coding Agents: Supercharging Downstream Consumer Dataset Migrations (Honk, Part 4)Engineering blog post, Spotify Engineering
A Spotify team rolled out 240 automated dataset migration PRs with its Claude Code based agent, work it estimated at about 10 engineering weeks by hand; it needed very detailed prompts, and it dropped one pipeline framework whose style varied too much between teams.
"We estimated that it would have taken around 10 engineering weeks of effort to complete these migrations manually."
"…we successfully rolled out 240 automated migration PRs using Fleetshift…"
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes
Positive
Chad Pytel, Developer and CEO at thoughtbot
Claude Code: Production ready code in a two-week sprintCompany blog post, thoughtbot blog
thoughtbot built production-quality Rails code for a client in a two-week sprint with Claude Code by giving it very small tasks, keeping a CLAUDE.md and reviewing every change before a human commit; sometimes it wrote tests that checked only the fake objects they set up.
"During a two-week sprint with a startup, we ended the intense engagement with true production-quality code…"
"Allowing Claude Code to make commits relinquishes this critical quality control step."
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes
Counted for Engineering leaders and teams.
Raised as a concern, 5
Mixed
linsomniac, Hacker News commenter
Comment on: Tokenmaxxing is dead, long live tokenmaxxingHacker News comment
At a company thrifty with AI spend, had a US$100 a month spend while trying to be its AI leader. Sees that Claude Code can do a lot of heavy lifting at work, while most colleagues stuck cost-maxing cannot imagine hitting session limits; says there's a case the cost-maxing hurt the company.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
Sergii Gorbachov, Staff Software Engineer at Slack
Agentic Testing: Where Agents Fit in the E2E Testing StackCompany engineering blog post, Slack Engineering
Ran more than 200 agentic end-to-end test runs at Slack through non-interactive Claude Code. Agent runs typically cost US$15 to 30 each and averaged about 5 to 11 minutes; says the cost may currently suit them better to debugging and exploratory testing than to high-frequency CI.
"Agent-driven runs were typically $15 to 30 per execution, compared to much cheaper traditional test runs."
"Due to the cost, agent-driven testing may currently be better suited for targeted debugging or exploratory testing than for high-frequency CI execution…"
Practitioner account, name and role on their organization's own page; structured test; counts toward Delivered outcomes
Mixed
Stella Laurenzo (stellaraccident), software engineer (employer unnamed in the issue)
[MODEL] Claude Code is unusable for complex engineering tasks with the Feb updatesGitHub issue
Reports from months of mined Claude Code logs on IREE projects using Opus that complex engineering quality degraded from February, forcing a shutdown of the concurrent agent fleet for single-session work. Says Claude has been good to them and hopes Anthropic can address it.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Reliability
Positive
Asher Moore, engineering leader at Airtasker
Rolling out Claude Code at AirtaskerEngineering blog post, Technology @ Airtasker (Medium)
Airtasker rolled Claude Code out across engineering via AWS Bedrock for data sovereignty, with managed settings, OpenTelemetry dashboards, and a custom spend tracker after CLI cost figures proved unreliable.
Practitioner account; ran the rollout; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
No opinion stated
Arunabh Dastidar, startup engineering lead
The big discount on intelligencePractitioner write-up
Five engineers who paid US$800 to US$1,000 a month together on Max and Pro plans paid about US$1,000 a day for the same work at API rates, Dastidar writes.
"After this, the bill was approximately $1000 per day for the same work. So in one fell swoop, the cost went up thirty times."
Discussed without concern, 6
Positive
jabart, Hacker News commenter
Comment on: Anthropic/OpenAI may be spending more than US$1000 for every US$100 you pay themHacker News comment
jabart's team measured usage with cc-usage and found costs near the monthly license on a multi-million-line codebase, even with jabart running 3 to 4 sessions at once.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Vincent Quigley, then Staff Engineer at Sanity
First attempt will be 95% garbage: A staff engineer's 6-week journey with Claude CodeEngineering blog post, Sanity Engineering
A Sanity engineer treats Claude Code like a junior developer with amnesia: three attempts per feature, Claude.md context, layered review, 2-3x faster shipping at roughly US$1000-1500 per month.
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes, Cost predictability and limits, Adoption and upkeep
Positive
Shrivu Shankar, leads AI developer tooling at Abnormal AI
How I Use Every Claude Code FeaturePersonal blog post, Shrivu's Substack (blog.sshh.io)
Describes how Abnormal AI uses Claude Code on its monorepo: a strictly maintained 13KB CLAUDE.md, clear-and-document restarts over auto-compaction, a hook that blocks commits until tests pass, and enterprise API keys for usage-based pricing.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Reliability, Operational control, Adoption and upkeep
Positive
Orta Therox, engineer at Puzzmo
6 Weeks of Claude CodeCompany engineering blog post, Puzzmo Blog
Cleared what Orta Therox calls years of tech-debt backlog solo in six weeks with Claude Code while working on the existing roadmap. Credits monorepos and boring, explicit technologies, and finds the change hard to see in pull request, commit and line counts.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Positive
Rory Bain, Product Engineer at incident.io
How we're shipping faster with Claude Code and Git WorktreesEngineering blog post, incident.io blog
incident.io engineers run several Claude Code agents in parallel via Git worktrees and plan mode; one US$8 session delivered an 18% faster API generation step, though architecture stays human-led.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control
Mixed
Alex Ghiculescu, co-founder of Tanda
Nobody Codes Here AnymoreNewsletter, Alex Ghiculescu's Newsletter (Substack)
Rolling out Cursor and Claude Code to about 40 developers, Tanda's co-founder finds Claude Code great at refactors but subtly wrong on bug fixes, estimating roughly 20% productivity gains.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Cost predictability and limits
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.Spending controls depend on the plan and billing route. Team and Enterprise admins can set spend limits for the organization, a group, or a member. Console workspaces have spend limits. A report of unexpected spend does not establish that spending controls are missing. Manage costs effectively
Open them in the recordCounted for Engineering leaders and teams.
Raised as a concern, 2
Mixed
Robat Williams, developer at Scott Logic
Choosing the right tool safety approach for coding agentsCompany engineering blog post, Scott Logic
Testing sandbox setups, Williams found that Claude Code's sandboxed Bash tool passes environment variables to the sandboxed process by default and can read the entire computer it runs on.
"Claude Code’s sandboxed Bash terminal tool by default exposes environment variables to the sandboxed process, and has read access to the entire computer it’s running on."
Practitioner account, self-described; direct use; counts toward Operational control
Negative
thewebguyd, Hacker News commenter
Comment on: Claude TagHacker News comment
Rolled Cowork out to all non-technical staff at work and calls Cowork/Claude Desktop frustrating, with no native audit log and all-or-nothing access for Claude Code and for some MCP connections such as QuickBooks Online. Wants governance features and per-member RBAC.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Discussed without concern, 8
Positive
Dan Guido, CEO at Trail of Bits
How we made Trail of Bits AI-native (so far)Company blog post, Trail of Bits Blog
Says Trail of Bits got everyone on Claude Code with supported configs, default sandboxing and a curated plugin marketplace. For certain clients where codebase and scope allow, auditors running fleets of agents went from about 15 bugs a week to 200.
"We got everyone on Claude Code, and we treat it like any other enterprise tool…"
"…we went from finding about 15 bugs a week to 200…"
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Positive
Brian Scanlan, engineer at Intercom
How we use Claude Code today at IntercomCompany blog post, Fin Ideas (Intercom)
Intercom turned Claude Code into an internal platform: 13 plugins, 100+ skills, hook-enforced PR workflow, guarded read-only production console, OpenTelemetry observability and auto-updated CLAUDE.md files.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Positive
Max Charas and Marc Bruggmann, Senior Staff and Principal Engineers at Spotify
Background Coding Agents: Context Engineering (Honk, Part 2)Engineering blog post, Spotify Engineering
Spotify moved its background migration agent to Claude Code after open source and homegrown agents struggled, finding end-state prompts, few tools and a verify tool give mergeable PRs.
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Operational control
Positive
Asher Moore, engineering leader at Airtasker
Rolling out Claude Code at AirtaskerEngineering blog post, Technology @ Airtasker (Medium)
Airtasker rolled Claude Code out across engineering via AWS Bedrock for data sovereignty, with managed settings, OpenTelemetry dashboards, and a custom spend tracker after CLI cost figures proved unreliable.
Practitioner account; ran the rollout; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Shrivu Shankar, leads AI developer tooling at Abnormal AI
How I Use Every Claude Code FeaturePersonal blog post, Shrivu's Substack (blog.sshh.io)
Describes how Abnormal AI uses Claude Code on its monorepo: a strictly maintained 13KB CLAUDE.md, clear-and-document restarts over auto-compaction, a hook that blocks commits until tests pass, and enterprise API keys for usage-based pricing.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Reliability, Operational control, Adoption and upkeep
Positive
Orta Therox, engineer at Puzzmo
6 Weeks of Claude CodeCompany engineering blog post, Puzzmo Blog
Cleared what Orta Therox calls years of tech-debt backlog solo in six weeks with Claude Code while working on the existing roadmap. Credits monorepos and boring, explicit technologies, and finds the change hard to see in pull request, commit and line counts.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Positive
Steve Sewell, Builder.io
How I use Claude Code (+ my best tips)Company blog post, Builder.io
After a year as a Cursor power user, Sewell moved his agent work to Claude Code, the only agent that has successfully updated an 18,000-line React component at Builder.io. He runs it with permission prompts skipped and calls the US$100 Max plan worth it.
"No AI agent has ever successfully updated this file except Claude Code."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Rory Bain, Product Engineer at incident.io
How we're shipping faster with Claude Code and Git WorktreesEngineering blog post, incident.io blog
incident.io engineers run several Claude Code agents in parallel via Git worktrees and plan mode; one US$8 session delivered an 18% faster API generation step, though architecture stays human-led.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.The operating-system sandbox for shell commands is off by default. Managed settings can disable the bypass and auto permission modes for an organization. Configure the sandboxed Bash tool
Open them in the recordStatements this group's sources agree on, then the ones close to agreement.
A maintained CLAUDE.md, shared rules, and written conventions make Claude Code's results more consistent across engineers and sessions.
24 supporting sourcesClaude Code is evaluated as an agent that carries out engineering tasks, with planning and review around it, and never as a completion tool.
18 supporting sourcesClose to agreement: each needs 14 supporting sources
The strongest reported workflows involve well-scoped repository tasks where Claude Code can inspect existing code, make the change, run tests, and iterate.
9 supporting sourcesAs usage grows, review and verification capacity becomes the limiting factor.
8 supporting sourcesOn self-directed work, Claude Code shortens the path from an idea to a working, tested implementation.
6 supporting sourcesQuestions the sources split on, with the count on each side.
Evidence remains mixed
Broader delegation, 9 sources
Some give the agent substantial scope, or run several sessions in parallel, once the objective and acceptance checks are explicit.
Tighter steering, 9 sources
Others review a plan first and work in small steps, so incorrect assumptions do not spread through larger changes.
Evidence remains mixed
Net time saved, 7 sources
Some report a clear gain after review, from doubled pull request throughput to hours saved a day.
Absorbed by review, 2 sources
Others find review, supervision, and cleanup absorb the time saved, or make the work slower than doing it by hand.
The three groups, then every practitioner by the context they work in, and how emphasis changes with scale.
Three groups drawn from the same sources. Choose a group to select whose experience counts. Within that group, each author counts once in the practitioner sentiment, with their latest view. Signals open on Engineering leaders and teams.
68 sources. These accounts focus on review capacity, team conventions, governance, cost visibility, delegation boundaries, and what changes when Claude Code moves from individual use into an engineering operating model.
70 sources. These accounts focus on raw capability, speed, experimentation, and how Claude Code performs in self-directed development workflows.
138 sources. This is the broadest read. It mixes settings, from solo work to company rollouts, that can lead to different conclusions.
138 sources from practitioners, by the context they work in. Filled rows count for this group.
Emphasizes Personal speed, usage limits, and context loss
Emphasizes Shipping velocity and how much to delegate
Emphasizes Review load and shared repository conventions
Emphasizes Where to delegate and how to measure it
Emphasizes Platform tooling, managed settings, and telemetry
What changes with scale. Individual builders emphasize personal speed and, increasingly, usage limits and cost. Team and organization accounts shift toward shared conventions, review load, managed settings, and telemetry.
Where this group's sources were published.
Counted for Engineering leaders and teams, from 68 sources. Change group
90 sources for this group, 22 of them documentation
Older evidence is retained when it still applies. Sources that depend on a changed plan, limit, or model stay in the history, marked Needs a recheck.
See sources by originWho each source is, how it knows, whether it still applies and what it counts toward.
Newest first, on its own page. Every count above opens its sources there.
Read all 160 sourcesAbout this read
A running review of what engineering teams report about Claude Code, from individual builders to company-wide rollouts. Each finding links to its sources, and each conflict shows both sides.
Counts cover only the sources GILD reviewed.
Choose a group to select whose experience counts. Within that group, each author counts once in the practitioner sentiment, with their latest view. GILD files each source in a group from how the account describes its author. A role or employer marked self-described is what the source states about themselves. Documentation checks facts for every group and never counts toward sentiment.
A source's date and whether it still applies are separate: a recent post can describe an old version, and an old one can still hold.
Flagged for a recheck, of the sources that take a side: leaders and teams 3 of 53, individual builders 12 of 57, all practitioners 15 of 110.
Each row opens its sources for the group you are reading.
| Finding | Engineering leaders and teams | Individual builders | All practitioners | Sources |
|---|---|---|---|---|
| Current readCurrent sentiment | 41 positive, 11 mixed, 1 negative | 23 positive, 22 mixed, 12 negative | 64 positive, 33 mixed, 13 negative | Sources |
| PraiseCodebase Understanding | 10 of 10 positive | 6 of 6 positive | 16 of 16 positive | Sources |
| PraiseMulti-Step Execution | 5 of 6 positive | 12 of 13 positive | 17 of 19 positive | Sources |
| PraiseLarge Refactors | 8 of 9 positive | 1 of 1 positive (under the minimum) | 9 of 10 positive | Sources |
| PraiseDeveloper Experience | 9 of 10 positive | 6 of 11 positive | 15 of 21 positive | Sources |
| PraiseValue for Money | 5 of 5 positive | 4 of 9 positive | 9 of 14 positive | Sources |
| PraiseEnterprise Administration | 4 of 5 positive | 1 of 3 positive (under the minimum) | 5 of 8 positive | Sources |
| PraiseStability | 0 of 1 positive (under the minimum) | 1 of 5 positive | 1 of 6 positive | Sources |
| ConcernUsage / Cost | 5 of 11 raise it | 14 of 20 raise it | 19 of 31 raise it | Sources |
| ConcernRequires Strong Review | 16 of 23 raise it | 10 of 17 raise it | 26 of 40 raise it | Sources |
| ConcernNeeds Close Steering | 3 of 4 raise it (under the minimum) | 4 of 5 raise it | 7 of 9 raise it | Sources |
| ConcernDay-to-Day Friction | 1 of 1 raise it (under the minimum) | 4 of 5 raise it | 5 of 6 raise it | Sources |
| ConcernReliability | 1 of 1 raise it (under the minimum) | 8 of 9 raise it | 9 of 10 raise it | Sources |
| ConcernContext Management | 6 of 12 raise it | 11 of 16 raise it | 17 of 28 raise it | Sources |
| ConcernInconsistent Results on Large Tasks | 6 of 7 raise it | 10 of 11 raise it | 16 of 18 raise it | Sources |
| ConcernPermissions & Safety | 2 of 10 raise it | 8 of 10 raise it | 10 of 20 raise it | Sources |
| ConsensusWritten conventions make results consistent | 24 supporting | 12 supporting (below threshold) | 36 supporting | Sources |
| ConsensusStrongest on scoped, repository-level work | 9 supporting (below threshold) | 9 supporting (below threshold) | 18 supporting (below threshold) | Sources |
| ConsensusEvaluated as an engineering agent | 18 supporting | 5 supporting (below threshold) | 23 supporting (below threshold) | Sources |
| ConsensusFaster on self-directed work | 6 supporting (below threshold) | 7 supporting (below threshold) | 13 supporting (below threshold) | Sources |
| ConsensusReview capacity becomes the constraint | 8 supporting (below threshold) | 0 supporting (under the minimum) | 8 supporting (below threshold) | Sources |
| DisagreementHow much implementation work should Claude Code complete before a human checkpoint? | Broader delegation 9, Tighter steering 9 | Broader delegation 9, Tighter steering 7 | Broader delegation 18, Tighter steering 16 | Sources |
| DisagreementOnce review and rework are counted, does Claude Code reduce total engineering time? | Net time saved 7, Absorbed by review 2 | Net time saved 5, Absorbed by review 4 | Net time saved 12, Absorbed by review 6 | Sources |
| DisagreementIs Claude Code better as the primary coding environment or as a complementary agent? | Primary environment 4, Paired with an IDE 0 (under the minimum) | Primary environment 5, Paired with an IDE 3 | Primary environment 9, Paired with an IDE 3 | Sources |
| Current readEvidence mix | 90 sources | 92 sources | 160 sources | Sources |
GILD's checks of claims against official documentation, kept apart from the source counts.
