What engineering leaders and their teams report about Cursor, with every finding tied to its evidence.
Read the findingsPractitioner sentiment, leaders and teams
+11 Mixed from 47 sources that take a side
How this figure is madeFrom 47 authors in 38 discussions. Use a task pilot to measure accepted work and review time.
Jan 2025 to Oct 2026.
162 sources from practitioners, by month and sentiment.
Team and leader accounts lean positive. Model choice draws the most praise, in 6 of 7 accounts. Cost draws the most concern, followed by uneven output and review load. Thirteen teams keep agent work in Cursor, while sixteen moved it to a separate agent, among them Applied Intuition.
Positive, mixed and negative sources
The score counts each source once, as the split does. See How the numbers work.
See the evidenceBelow: 3 praise themes, 4 concerns, 4 close to agreement and 2 disagreements.
Counted for Engineering leaders and teams, from 73 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, 6
Positive
ziml77, Hacker News commenter, self-described
Comment on: Graphite is joining CursorHacker News comment
For their business, Cursor's value is many models under one contract with data privacy terms that cover whichever model a developer picks.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Mixed
zxspectrum1982, Hacker News commenter
Comment on: Our decision on Cursor following its acquisition by SpaceXHacker News comment
Loves Cursor at work, but since the Cursor Token Rate their company tells developers to use Claude and Codex directly, so the commenter uses Cursor-native Grok models almost exclusively.
not counted, repeats a stronger account
Positive
joshgachnang, Hacker News commenter
Comment on: SpaceX to buy Cursor for US$60BHacker News comment
Just set their team up on a Cursor enterprise plan: cloud agents with computer use, the best AI reviewer the commenter has tested in Bugbot, and Composer 2.5 with Opus for planning.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Positive
phsource, Hacker News commenter
Comment on: Show HN: Boxes.dev: ditch localhost; run Claude Code and Codex in the cloudHacker News comment
Their company runs on Cursor and picks models per job: Kimi for easy fixes, Opus or Sonnet for plans, Composer 2.5 and smaller models for fast implementation.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
simplyluke, Hacker News commenter
Comment on: Cursor 3Hacker News comment
A daily user in a large production codebase delegates simpler Linear tasks to cloud agents and picks them up in the editor, and values not being tied to one model provider.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Positive
NitpickLawyer, Hacker News commenter, self-described
Comment on: Composer: Building a fast frontier model with RLHacker News comment
Their team runs its own tiered task suite on each new model and finds cheaper models such as Cursor's fast model good enough for some tasks while reserving larger ones for others.
Practitioner account, pseudonymous; opinion; not counted, takes no position
Mixed or negative, 1
Negative
bushido, Hacker News commenter who sets tool policy at their company
Comment on: Shall I implement it? NoHacker News comment
Barred Cursor at their company after its automatic model and effort selection moved to weaker models mid-task and produced bad results.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
GILD's fact checks, which don't count as sources
Checked Oct 28, 2026.Each X Subscription plan includes one weekly pool for Grok and Composer, and third-party models bill at API list price from prepaid credit from the first request. Legacy Pro, Pro+, and Ultra plans keep two monthly pools, Cursor Models and Other Models. Model choice therefore decides how fast included usage runs out and what lands on the bill. Models & Pricing
Open them in the recordCounted for Engineering leaders and teams.
Positive, 4
Positive
Manoj Vaitla, senior engineer at Salesforce
How AI Test Automation Cut Developer Productivity Bottlenecks by 30% at ScaleCompany engineering blog post, Salesforce Engineering
Manoj Vaitla's Salesforce team built its test failure triage agent with Cursor in 4 to 6 weeks, work they expected to take months, and credits Cursor's codebase-aware suggestions. The agent cut failure resolution time by 30%.
"What would have taken months to build, the team delivered in just 4-6 weeks. Cursor’s deep integration with the codebase provided immediate, contextually relevant suggestions during the development of the TF-Triage Agent."
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes
Positive
Akshay Misra and Karmi Nguyen, engineering blog authors at Grab
Cursor at Grab: Adoption and impactCompany engineering blog post, Grab Tech
Grab reports that 98% of its tech staff became monthly Cursor users and about 75% use it weekly, with about half of suggestions accepted and over a third of merge requests using Cursor. Engineers say tasks that took a full day now take hours, helped by preconfigured rules for Grab's conventions.
"Qualitative feedback confirms what the adoption numbers suggest: tasks that took a full day to complete now take hours."
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Adoption and upkeep
Positive
Kristin Foss, VP of Software Engineering for Customer Centric Engineering at Salesforce
How AI Tools Cut Customer Escalation Time: From Days of Manual Work to MinutesCompany engineering blog post, Salesforce Engineering
Kristin Foss's 180-person Salesforce team uses Cursor, Windsurf and Claude CLI to build working reproductions of customer issues from internal source code, cutting reproduction work that took hours or days down to minutes.
"…our CCE engineers now leverage Cursor, Windsurf, and Claude CLI. These AI-powered development tools allow them to quickly create working reproductions that mirror customer problems using internal source code, drastically speeding up our debugging workflows."
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes
Positive
Daniel Koch, vice president of engineering at Jit
How Jit overcame developer resistance to shift to CursorArticle, LeadDev
Jit's VP of engineering moved the whole engineering team from JetBrains to Cursor and ran workshops on Cursor rules. Once the rules were written the agent followed the codebase, and two months in the team measured 1.8x to 3x faster delivery of new features.
"…once we defined the right [Cursor] rules, we started getting better results. The agents could adapt, understand our codebase, and behave much more intelligently compared to working without any guidance."
Practitioner account, self-described; second-hand; not counted until substantiated
Mixed or negative, 3
Mixed
bunderbunder, Hacker News commenter
Comment on: Two things LLM coding agents are still bad atHacker News comment
Reviewing a feature written with Cursor, the commenter judged only about 200 lines necessary because it ignored the team's utility library. Approved it since hunting the right functions would take all day.
Practitioner account, pseudonymous; second-hand; not counted, repeats a stronger account
Positive
Scott Yancey, EVP of Software Engineering at Salesforce
Building Sustainable Enterprise AI Adoption: Cultural Strategies That Achieved 95% Developer EngagementEngineering blog Q&A, Salesforce Engineering Blog
Cursor could not handle Salesforce's monolith of hundreds of thousands of files, so the team scoped workspaces to team-owned modules and adopted it for the 80% of work it fits.
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes, Adoption and upkeep
Negative
eranation, Hacker News commenter
Comment on: Cursor IDE support hallucinates lockout policy, causes user cancellationsHacker News comment
Moved their whole team to Cursor as an enterprise customer and is considering canceling over nonexistent support and bugs such as creating new files where an edit belongs in a monorepo.
Practitioner account, pseudonymous; ran the rollout; counts toward Delivered outcomes
Counted for Engineering leaders and teams.
Positive, 3
Positive
benbayard, Hacker News commenter
Comment on: Cursor CLIHacker News comment
Cursor has been a workhorse at their company for months, while Claude Code looped and deleted files in their codebase without completing a meaningful ticket.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Vivienne Wei, with Geeta Shankar and Shiva Nimmagadda, Salesforce Engineering
How Salesforce Engineering Operationalized AI Productivity at ScaleEngineering blog post, Salesforce Engineering Blog
Salesforce made Cursor a default coding agent, reached over 90% engagement across six engineering clouds in three weeks, and upgraded 475 engineers who exceeded default token limits.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Cost predictability and limits
Positive
Lakshmi Dontamsetti, Senior Manager of Software Engineering for Mobile Performance at Salesforce
How Cursor AI Slashed Dashboard Migration Time 75% Across 240 QueriesEngineering blog Q&A, Salesforce Engineering Blog
Dontamsetti's team used Cursor and an MCP server to move Splunk dashboards to Tableau, backing up about 240 queries across 20 dashboards, and reports engineering time cut 75%. Explicit rules and a schema mapping document fixed semantically wrong early queries; 20 to 30 percent needed human oversight.
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes
Mixed or negative, 2
Mixed
Alex Ghiculescu, co-founder of Tanda
Nobody Codes Here Anymore: Notes on rolling out Cursor and Claude CodeNewsletter, Alex Ghiculescu's Newsletter (Substack)
Offering about 40 developers Cursor or Claude Code, Ghiculescu saw slightly higher Cursor adoption and says that, from what he has seen, Cursor "seems to be content making smaller and more self contained changes". Sticks with Claude Code to keep Sublime Text; set up Cursor rules and estimates being about 20% more productive thanks to agents.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Mixed
Steve Sewell, co-founder and CEO of Builder.io
How I use Claude Code (+ my best tips)Company blog post, Builder.io blog
A Cursor power user for over a year, Sewell moved agent work to Claude Code, saying no AI agent but Claude Code has updated Builder's 18,000-line React component and Cursor needs more babysitting. Still uses Cursor for Command+K and tab completions.
"I still use Cursor for quick Command+K completions and tab completions. But the agent sidebar? I only touch it when Claude is down."
Vendor material, self-described; direct use; not counted, from a competitor's side
Every concern practitioners raise, most raised first.
Counted for Engineering leaders and teams.
Raised as a concern, 11
Mixed
zxspectrum1982, Hacker News commenter
Comment on: Our decision on Cursor following its acquisition by SpaceXHacker News comment
Loves Cursor at work, but since the Cursor Token Rate their company tells developers to use Claude and Codex directly, so the commenter uses Cursor-native Grok models almost exclusively.
not counted, repeats a stronger account
Negative
Jovan (jovanvel), early Cursor power user who brought colleagues onto Cursor
Reply in: Legacy Individual Plan, Max Mode RequiredCursor forum post
An early power user who brought over 30 colleagues onto Cursor calls the mid-subscription Max Mode requirement on legacy plans a betrayal of the terms the poster renewed under.
Practitioner account, pseudonymous; direct use; counts toward Cost predictability and limits
Mixed
sanex, Hacker News commenter, self-described
Comment on: SpaceX to buy Cursor for US$60BHacker News comment
Their whole team used Cursor for months and liked it. Spend of US$500 to 1,000 a month per person led the business to switch them to Claude Code for its cost controls.
Practitioner account, pseudonymous; direct use; counts toward Cost predictability and limits
Negative
dirtbag__dad, Hacker News commenter
Comment on: Cursor 3Hacker News comment
Was spending around US$2k a week on premium models in Cursor when their team unanimously chose hiring another engineer over more Cursor. Switched to Claude Code Max, equally prolific at 1/10th the price, and calls Cursor's pricing model problematic.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Negative
mc_escher, Hacker News commenter, self-described
Comment on: Cursor 3Hacker News comment
Their team stopped using Cursor after US$2,500 of unauthorized weekend usage, citing no 2FA and risky defaults such as admin-only usage settings being off.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Negative
Jcampuzano2, Hacker News commenter
Comment on: Cursor 3Hacker News comment
Their company calculated enterprise renewal pricing rising four to eight times under usage-based terms with a per-token surcharge, and committed to far less Cursor spend.
Practitioner account, pseudonymous; ran the rollout; counts toward Cost predictability and limits
Negative
primex, Hacker News commenter
Cursor billed us US$450 for a seat that existed for secondsHacker News thread
Their team was billed about US$450 for a full annual Teams seat added by mistake and removed within seconds, with no warning in the UI, and support called it expected.
Practitioner account, pseudonymous; ran the rollout; counts toward Cost predictability and limits
Negative
songcj, admin of an annual Cursor Team plan
Team Seat Management Best Practices: How Do You Handle Member Changes?Cursor forum post
A team admin found 66 of 86 annual seats still held by removed members, with only 13 active, and asks how to avoid paying for seats that cannot be reassigned.
Practitioner account, pseudonymous; ran the rollout; not counted, takes no position
Mixed
valentinoPereira, member of a team on a Cursor team plan
Reply in: Anyone here using cursor with a team? what's been the hardest part?Cursor forum post
Their team uses Cursor rules heavily but still finds model output unpredictable, and runs out of allocated credits, keeping a Google AI Pro plan as backup.
Practitioner account, pseudonymous; direct use; counts toward Cost predictability and limits
Mixed
Flavien, Lead Frontend Developer
Why we're moving from Cursor to Claude Code (and why you should too)Personal blog post, 56kode
A lead frontend developer says Cursor's business model isn't ideal for teams, since Opus is expensive and overage is shared at the organization level, and parallel agents saturated RAM across IDE instances. Moved to Claude Code after shipping a personal project with coding agents.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Cost predictability and limits
Negative
Stergios (StergiosDikos), developer on a team that accesses Claude through AWS Bedrock
Reply in: Stop Shipping Unstable BuildsCursor forum post
Their team reports bugs and poor support: enabling AWS Bedrock for Claude stops every other model, and Cursor stuck in Max mode doubles their cost.
Practitioner account, pseudonymous; direct use; counts toward Operational control, Cost predictability and limits
Discussed without concern, 5
Positive
bazzargh, Hacker News commenter
Comment on: SpaceX to buy Cursor for US$60BHacker News comment
Their team handles the AI bills: Cursor on Auto averaged 3.9 cents a request under the old per-seat contract and 39 cents under the pooled one, with Claude models costing dollars.
Practitioner account, pseudonymous; ran the rollout; counts toward Operational control
Positive
Mark Harrison, member of a team on Cursor Pro
Reply in: Anyone here using cursor with a team? what's been the hardest part?Cursor forum post
Their team keeps agent rules in the repo through AGENTS.md, nearly dropped Cursor in late 2025 but renewed after it improved, and stays near 80% of allowance by planning first on auto.
Practitioner account, pseudonymous; ran the rollout; counts toward Cost predictability and limits
Positive
Vivienne Wei, with Geeta Shankar and Shiva Nimmagadda, Salesforce Engineering
How Salesforce Engineering Operationalized AI Productivity at ScaleEngineering blog post, Salesforce Engineering Blog
Salesforce made Cursor a default coding agent, reached over 90% engagement across six engineering clouds in three weeks, and upgraded 475 engineers who exceeded default token limits.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Cost predictability and limits
Positive
Peter Eysermans, engineer at madewithlove
Automatic pull request reviewing with Cursor's BugbotEngineering blog post, madewithlove blog
Has used Bugbot since the beta; in a month it caught around ten real bugs, with occasional false positives. Calls US$40/month on the more expensive end and says Bugbot can offer real value for small teams with limited review time.
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 Anymore: Notes on rolling out Cursor and Claude CodeNewsletter, Alex Ghiculescu's Newsletter (Substack)
Offering about 40 developers Cursor or Claude Code, Ghiculescu saw slightly higher Cursor adoption and says that, from what he has seen, Cursor "seems to be content making smaller and more self contained changes". Sticks with Claude Code to keep Sublime Text; set up Cursor rules and estimates being about 20% more productive thanks to agents.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control, Cost predictability and limits
GILD's fact checks, which don't count as sources
Checked Oct 28, 2026.On current Teams plans, additional usage draws from a prepaid team balance, and admins set a monthly team limit, a default per-member limit, and per-member overrides. Legacy Teams seats keep on-demand usage on by default with a team-wide spend limit. A report of unexpected spend does not establish that spending controls are missing. Team Pricing
Open them in the recordCounted for Engineering leaders and teams.
Negative
bushido, Hacker News commenter who sets tool policy at their company
Comment on: Shall I implement it? NoHacker News comment
Barred Cursor at their company after its automatic model and effort selection moved to weaker models mid-task and produced bad results.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
valentinoPereira, member of a team on a Cursor team plan
Reply in: Anyone here using cursor with a team? what's been the hardest part?Cursor forum post
Their team uses Cursor rules heavily but still finds model output unpredictable, and runs out of allocated credits, keeping a Google AI Pro plan as backup.
Practitioner account, pseudonymous; direct use; counts toward Cost predictability and limits
Negative
interestpiqued, Hacker News commenter
Comment on: Claude Code is suddenly everywhere inside MicrosoftHacker News comment
Their firmware team had a Cursor honeymoon before it was sidelined: the agent usually got a couple of attempts, then the engineers solved the problem themselves.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Negative
Jeremy Crane, founder and CEO of PocketOS
An AI Agent Just Destroyed Our Production Data. It Confessed in Writing.X post
PocketOS's founder reports that Cursor's agent, running Claude Opus 4.6 under explicit project safety rules, deleted the company's production database and its volume backups in one Railway API call. He concludes that rules in a system prompt cannot be the only guard against destructive actions.
"Cursor running Anthropic's flagship Claude Opus 4.6, deleted our production database and all volume-level backups in a single API call to Railway…"
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Reliability, Operational control
Mixed
Tripti Sheth, Senior Director of Software Engineering, Monitoring Cloud Notification Management, Salesforce
Building AI-Powered Migration Tools: Compressing 4 Sprints Into 3 Days with Cursor, Windsurf, and ClaudeCompany engineering blog post, Salesforce Engineering
Sheth's Salesforce team began its migration tool in Cursor with plain prompts and no rules, which produced unmanageable code adding 1,000 lines per simple task. The team then used rules and memories in Windsurf and a shared CLAUDE.md file in Claude.
"Initially, we started with Cursor, using straightforward prompts without any rules or structured knowledge. This approach led to unmanageable code, where each simple task added 1,000 lines with no modularization, reuse, or abstraction."
Practitioner account, name and role on their organization's own page; direct use; counts toward Delivered outcomes
Negative
Adrian (QTom), developer whose company provides AI coding tools
Reply in: 2.0, A step in the wrong directionCursor forum post
While Cursor 2.0 performed poorly, their company gave the team Claude Max, and a task the poster had fought in Cursor for a week took a single day in Claude.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Mixed
hallh, Hacker News commenter, self-described
Comment on: Notes on rolling out Cursor and Claude CodeHacker News comment
On a newer React project, fast lint and test runs in Cursor were essential because models kept using outdated patterns. With Gemini they got a polished UI only with a lot of guidance.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Counted for Engineering leaders and teams.
Raised as a concern, 6
Positive
Antoine Cormouls, developer at a company using Cursor on a large monorepo
Reply in: Anyone here using cursor with a team? what's been the hardest part?Cursor forum post
Their company runs Plan Mode, agent implementation against tests, developer review in the IDE, automated review and a final human review, since code is easy to produce and review is the bottleneck.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Negative
cmiles74, Hacker News commenter
Comment on: We mourn our craftHacker News comment
Junior engineers on their team using Cursor and Claude have checked in small breaking query changes, and AI tooling has nearly doubled the time the commenter spends on pull request review.
Practitioner account, pseudonymous; second-hand; not counted, repeats a stronger account
Mixed
tdu01, Hacker News commenter who leads a team
Comment on: Show HN: How well do you use Claude Code?Hacker News comment
Their team is very productive with Cursor using different models per task, but senior engineers cannot keep up reviewing the volume, and the commenter worries about pace versus quality.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Positive
Thoughtworks Technology Radar, April 2026 edition
Cursor | Technology Radar | ThoughtworksTechnology Radar assessment, Thoughtworks
Thoughtworks puts Cursor in its Adopt ring as a default choice for its delivery teams beside Claude Code. Many of its developers prefer supervising the agent in the IDE to review and refine plans.
"…consistently appearing alongside Claude Code as a default choice for our delivery teams. It has matured into a comprehensive agentic environment with features such as plan mode, hooks and subagents. While terminal-based agents remain popular, many of our developers find that supervising an agent within an IDE provides a richer experience for reviewing and refining plans before execution."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Mixed
bunderbunder, Hacker News commenter
Comment on: Two things LLM coding agents are still bad atHacker News comment
Reviewing a feature written with Cursor, the commenter judged only about 200 lines necessary because it ignored the team's utility library. Approved it since hunting the right functions would take all day.
Practitioner account, pseudonymous; second-hand; not counted, repeats a stronger account
Mixed
Alex Ghiculescu, co-founder of Tanda
Nobody Codes Here Anymore: Notes on rolling out Cursor and Claude CodeNewsletter, Alex Ghiculescu's Newsletter (Substack)
Offering about 40 developers Cursor or Claude Code, Ghiculescu saw slightly higher Cursor adoption and says that, from what he has seen, Cursor "seems to be content making smaller and more self contained changes". Sticks with Claude Code to keep Sublime Text; set up Cursor rules and estimates being about 20% more productive thanks to agents.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Discussed without concern, 3
Positive
fourseventy, Hacker News commenter
Comment on: Today I've made the difficult decision to reduce the size of Coinbase by ~14%Hacker News comment
Their company's mostly non-technical marketing team uses Cursor web agents to change the website, with the agents opening pull requests the tech team reviews before merge.
Practitioner account, pseudonymous; second-hand; not counted, repeats a stronger account
Positive
znq, SIROC venture studio
Comment on: Ask HN: Has your whole engineering team gone big into AI coding? How's it going?Hacker News comment
At an 18-person venture studio pushing Cursor and Claude Code adoption, well-specified prompts turned a 20-hour effort into about 3 hours, while weak specs and no verification produce broken systems.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Positive
Lakshmi Dontamsetti, Senior Manager of Software Engineering for Mobile Performance at Salesforce
How Cursor AI Slashed Dashboard Migration Time 75% Across 240 QueriesEngineering blog Q&A, Salesforce Engineering Blog
Dontamsetti's team used Cursor and an MCP server to move Splunk dashboards to Tableau, backing up about 240 queries across 20 dashboards, and reports engineering time cut 75%. Explicit rules and a schema mapping document fixed semantically wrong early queries; 20 to 30 percent needed human oversight.
Practitioner account, name and role on their organization's own page; ran the rollout; counts toward Delivered outcomes
Counted for Engineering leaders and teams.
Raised as a concern, 3
Mixed
rob74, Hacker News commenter
Comment on: SpaceX to buy Cursor for US$60BHacker News comment
Their company went all in on Cursor and disabled JetBrains AI, until a large group of developers pushed IT to enable plugins that run Cursor in JetBrains via ACP.
Practitioner account, pseudonymous; direct use; counts toward Adoption and upkeep
Mixed
saltyoldman, Hacker News commenter
Comment on: Cursor 1.7Hacker News comment
Their company standardized on Cursor; the commenter likes its tab completion but finds the accept and undo flow redundant with Git and resents a fork that wipes their Python extension stack.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
Alex Ghiculescu, co-founder of Tanda
Nobody Codes Here Anymore: Notes on rolling out Cursor and Claude CodeNewsletter, Alex Ghiculescu's Newsletter (Substack)
Offering about 40 developers Cursor or Claude Code, Ghiculescu saw slightly higher Cursor adoption and says that, from what he has seen, Cursor "seems to be content making smaller and more self contained changes". Sticks with Claude Code to keep Sublime Text; set up Cursor rules and estimates being about 20% more productive thanks to agents.
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Discussed without concern, 3
No opinion stated
Abhishek Tibrewal, Uber, presenting with Ty Smith at DX Annual
Uber's journey of measuring AI impact on developer productivityConference session recording, Engineering Enablement podcast (DX)
Says Uber's adoption data showed its iOS developers were not picking up AI tools because Xcode lagged in AI capabilities, so Uber reprioritized Swift LSP work and made Cursor available to them.
"…the IS developers were not picking up these tools, and so we figured out the X code was lagging behind in AI capabilities. And so we reprioritized our efforts in Swift LSP and made Cursor available to them. Adoption data told us exactly…"
Practitioner account; ran the rollout; not counted, takes no position
Negative
trial3, Hacker News commenter
Comment on: SpaceX to buy Cursor for US$60BHacker News comment
Their employer, a large enterprise customer, dropped Cursor for Claude; after eight months on Cursor as their main IDE the commenter switched easily and no coworker misses it.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Thoughtworks Technology Radar, April 2026 edition
Cursor | Technology Radar | ThoughtworksTechnology Radar assessment, Thoughtworks
Thoughtworks puts Cursor in its Adopt ring as a default choice for its delivery teams beside Claude Code. Many of its developers prefer supervising the agent in the IDE to review and refine plans.
"…consistently appearing alongside Claude Code as a default choice for our delivery teams. It has matured into a comprehensive agentic environment with features such as plan mode, hooks and subagents. While terminal-based agents remain popular, many of our developers find that supervising an agent within an IDE provides a richer experience for reviewing and refining plans before execution."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Adoption and upkeep
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.Cursor's agent now runs inside JetBrains IDEs through the Agent Client Protocol on paid plans. VS Code extensions, themes, settings, and keybindings import in one step. JetBrains
Open them in the recordStatements this group's sources agree on, then the ones close to agreement.
Close to agreement: each needs 15 supporting sources
Written rules, in .cursor/rules or AGENTS.md, make Cursor's output more consistent across engineers and sessions.
6 supporting sourcesPicking the model for each task, where one default would serve every request, changes both the result and how fast usage runs out.
6 supporting sourcesThe strongest reported workflows break work into scoped tasks with tests or clear checks the agent can run.
6 supporting sourcesAs agent output grows, review and verification capacity becomes the limiting factor.
6 supporting sourcesQuestions the sources split on, with the count on each side.
Evidence remains mixed
Cursor's own agent, 13 sources
Some run agents in Cursor's editor or its cloud agents, valuing model choice and an editor for reviewing diffs.
A separate agent, 16 sources
Others keep Cursor or VS Code as the editor and move agent work to Claude Code or Codex, often for cost.
Evidence remains mixed
Net time saved, 11 sources
Some report a clear gain after review, from a 75% cut in migration time to a personal estimate of 20%.
Absorbed by review, 3 sources
Others find review, rework, or fighting the tool absorbs the gain, or makes the work slower.
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.
73 sources. Leaders' and teams' accounts cover shared rules, review capacity, spend and renewal decisions, and whether agent work stays in Cursor once it moves from individual use into a team.
89 sources. Builders' accounts cover Tab completion, model choice, usage limits, and how Cursor performs in self-directed development.
162 sources. All accounts together give the broadest market read. They mix individual work with team rollouts, and the two can reach very different conclusions.
162 sources from practitioners, by the context they work in. Filled rows count for this group.
Emphasizes Model choice and usage limits
Emphasizes Spend per seat and where agent work runs
Emphasizes Shared rules and renewal cost
Emphasizes Where Cursor fits in the tool mix
Emphasizes Rollout scope and contract renewal
What changes with scale. Individual builders write most about picking models and running into usage caps. Team and organization accounts focus on shared rules, review load, renewal cost, and whether agents stay in Cursor.
Where this group's sources were published.
Counted for Engineering leaders and teams, from 73 sources. Change group
94 sources for this group, 21 of them documentation
Older evidence stays when it still applies. Sources tied to a changed plan, price, limit, or model remain 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 183 sourcesAbout this read
A running review of what engineering teams report about Cursor, 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 7 of 47, individual builders 11 of 70, all practitioners 18 of 117.
Each row opens its sources for the group you are reading.
| Finding | Engineering leaders and teams | Individual builders | All practitioners | Sources |
|---|---|---|---|---|
| Current readCurrent sentiment | 19 positive, 14 mixed, 14 negative | 30 positive, 13 mixed, 27 negative | 49 positive, 27 mixed, 41 negative | Sources |
| PraiseModel Choice | 6 of 7 positive | 14 of 18 positive | 20 of 25 positive | Sources |
| PraiseTab Completion | 2 of 2 positive (under the minimum) | 8 of 11 positive | 10 of 13 positive | Sources |
| PraiseAgent Execution | 3 of 5 positive | 6 of 7 positive | 9 of 12 positive | Sources |
| PraiseCloud Agents | 4 of 4 positive (under the minimum) | 4 of 5 positive | 8 of 9 positive | Sources |
| PraiseCodebase Understanding | 4 of 7 positive | 5 of 8 positive | 9 of 15 positive | Sources |
| PraiseValue for Money | 0 of 2 positive (under the minimum) | 7 of 11 positive | 7 of 13 positive | Sources |
| PraiseSpeed | 2 of 2 positive (under the minimum) | 6 of 8 positive | 8 of 10 positive | Sources |
| ConcernUsage / Cost | 11 of 16 raise it | 14 of 19 raise it | 25 of 35 raise it | Sources |
| ConcernRequires Strong Review | 6 of 9 raise it | 8 of 10 raise it | 14 of 19 raise it | Sources |
| ConcernNeeds Close Steering | 2 of 2 raise it (under the minimum) | 8 of 10 raise it | 10 of 12 raise it | Sources |
| ConcernInconsistent Quality | 7 of 7 raise it | 7 of 8 raise it | 14 of 15 raise it | Sources |
| ConcernReliability & Performance | 4 of 4 raise it (under the minimum) | 8 of 8 raise it | 12 of 12 raise it | Sources |
| ConcernEditor Lock-in | 3 of 6 raise it | 4 of 5 raise it | 7 of 11 raise it | Sources |
| ConsensusWritten rules make results consistent | 6 supporting (below threshold) | 4 supporting (under the minimum) | 10 supporting (below threshold) | Sources |
| ConsensusChoosing the model per task decides quality and cost | 6 supporting (below threshold) | 8 supporting (below threshold) | 14 supporting (below threshold) | Sources |
| ConsensusStrongest on scoped, testable work | 6 supporting (below threshold) | 2 supporting (under the minimum) | 8 supporting (below threshold) | Sources |
| ConsensusFaster on self-directed work | 0 supporting (under the minimum) | 6 supporting (below threshold) | 6 supporting (below threshold) | Sources |
| ConsensusReview capacity becomes the constraint | 6 supporting (below threshold) | 0 supporting (under the minimum) | 6 supporting (below threshold) | Sources |
| DisagreementShould agent work run in Cursor, or beside it in a separate agent? | Cursor's own agent 13, A separate agent 16 | Cursor's own agent 7, A separate agent 15 | Cursor's own agent 20, A separate agent 31 | Sources |
| DisagreementOnce review and rework are counted, does Cursor reduce total engineering time? | Net time saved 11, Absorbed by review 3 | Net time saved 0, Absorbed by review 4 (under the minimum) | Net time saved 11, Absorbed by review 7 | Sources |
| DisagreementHow much should Cursor's agent complete before a human checkpoint? | Broader delegation 1, Tighter steering 2 (under the minimum) | Broader delegation 4, Tighter steering 13 | Broader delegation 5, Tighter steering 15 | Sources |
| Current readEvidence mix | 94 sources | 110 sources | 183 sources | Sources |
GILD's checks of claims against official documentation, kept apart from the source counts.
