What engineering leaders and their teams report about Cline, with every finding tied to its evidence.
Read the findingsPractitioner sentiment, all practitioners
+39 Positive from 73 sources that take a side
How this figure is madeFrom 70 authors in 67 discussions. Use a task pilot to measure accepted work and review time.
Jan 2025 to Oct 2026.
94 sources from practitioners, by month and sentiment.
Across public accounts the read is positive. Model and provider choice is the most praised capability. Spending tokens on your own API key is the most frequent concern, followed by update regressions and results that depend on the model.
Positive, mixed and negative sources
The score counts them as 40 positive, 17 mixed and 13 negative votes, one per author. See How the numbers work.
See the evidenceBelow: 4 praise themes, 5 concerns, 5 close to agreement and 3 disagreements.
Counted for All practitioners, from 94 sources. Change group
Every capability practitioners describe positively, ranked by how many sources say so. Open one to read who said it.
Counted for All practitioners.
Positive
zenoprax, Hacker News commenter
Comment on: Cursor 3Hacker News comment
Switched from Cursor to VSCodium with Cline and OpenRouter; zenoprax uses any model, currently Step 3.5 Flash, for pennies, with no Cline account needed.
Practitioner account, pseudonymous; direct use; counts toward Adoption and upkeep
Positive
masto, Hacker News commenter
Comment on: I came back from Cursor to VS CodeHacker News comment
Prefers VS Code with Cline; likes the wide choice of models and payment options, and pay-as-you-go suits bursty use.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
syspec, Hacker News commenter
Comment on: Gemini 3Hacker News comment
With work access to Gemini 2.5 Pro and Claude Sonnet 4.5, syspec compares them in the same repository through Cline's model dropdown and prefers Claude's leaner code.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
esperent, Hacker News commenter
Comment on: Claude Code 2.0Hacker News comment
Found Claude Code very expensive and no better than Cline on similar tasks, so esperent stays with Cline and switches APIs by price and performance.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
efitz, Hacker News commenter, self-described
Comment on: OpenAI's Windsurf deal is off, and Windsurf's CEO is going to GoogleHacker News comment
Calls Cline and Roo Code fantastic as normal VS Code extensions, notes they occasionally lose the terminal connection, and prefers bringing a personal key over an IDE middleman.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
chaboud, Hacker News commenter
Comment on: OpenAI o3 and o4-miniHacker News comment
Gets great results from Cline with Sonnet 3.7 on AWS Bedrock for smaller projects in popular languages, but warns purely prompt-built projects degrade as context windows saturate.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
danielbln, Hacker News commenter, self-described
Comment on: Roo or Cline? We're building a supersetHacker News comment
Finds Cline better than Claude Code because it runs terminal commands, uses LSP and IDE diagnostics, and works with Gemini 2.5 alongside Anthropic models.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Addy Osmani, engineer and author
Why I use Cline for AI EngineeringNewsletter, Elevate (addyo.substack.com)
After extensively testing Cursor, WindSurf and GitHub Copilot, Osmani finds Cline uniquely valuable for model flexibility, Plan/Act, checkpoints, MCP extensibility and cost accounting during a session, while noting token-based pricing requires attention to model selection.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Marha01, Hacker News commenter
Comment on: Fable 5 is BackHacker News comment
Uses Cline's automatic model switch between Plan and Act: Opus for planning, Sonnet for implementation.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
DrAwdeOccarim, Hacker News commenter
Comment on: A guide to local coding modelsHacker News comment
Powers Cline with Devstral Small 2 served by vLLM on a DGX Spark; it works but slowly, and DrAwdeOccarim is impressed by it.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
Fergusonb, Hacker News commenter, self-described
Comment on: Claude Haiku 4.5Hacker News comment
Says GPT 5 in Cline reads whatever context it is given and laser-targets the required changes; on High, with enough relevant context, it usually one-shots the solution. Finds it extremely slow but still uses it on anything nuanced.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
chisleu, Hacker News commenter
Comment on: Claude Opus 4.1Hacker News comment
Uses Opus or Gemini 2.5 Pro for Plan mode and Sonnet for Act mode in Cline, finding Opus better at architectural problems.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
reilly3000, Hacker News commenter
Comment on: Tokens are getting more expensiveHacker News comment
Likes Cline with OpenRouter for its per-task token count, context size and live price, which help decide which model gets which work.
Practitioner account, pseudonymous; direct use; counts toward Cost predictability and limits
Positive
Lorenz Hofmann-Wellenhof, engineer
How a Real AI Agent Works Under the HoodNewsletter post, The Growing Engineer (Substack)
After a couple of months on Cline, Hofmann-Wellenhof keeps Cursor for autocomplete and codebase questions but prefers Cline whenever an agent should work more freely on the codebase.
"I still use Cursor often, mainly for its autocomplete model (e.g., via the Tab key) and for asking questions about the codebase. But when it comes to letting an agent operate more freely on the codebase, I prefer Cline."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control
Positive
theshrike79, Hacker News commenter
Comment on: Perverse incentives of vibe codingHacker News comment
Settled on Cline through OpenRouter, with Gemini 2.5 for Plan mode and Claude 3.7 for Act mode, after finding Claude Code and Aider overreached.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
Flux159, Hacker News commenter
Comment on: Ask HN: How much better are AI IDEs vs. copy pasting into chat apps?Hacker News comment
Moved to Cline with Gemini 2.5 Pro and a .clinerules file after Claude Code cost US$3 to US$5 per task. It runs commands and lint checks and costs little on a personal key.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
CalmStorm, Hacker News commenter
Comment on: OpenAI looked at buying Cursor creator before turning to WindsurfHacker News comment
Tried Cline and Cursor and finds Cline with DeepSeek V3 comparable to Cursor and more cost-effective.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Addy Osmani, engineer and author
What's new in @cline? (X post)X post
Osmani plans in Cline with Gemini or DeepSeek and executes with Sonnet in Act mode, a split he uses to keep token costs down, and recommends Cline to Cursor and Windsurf users.
"I use Plan mode with Gemini or DeepSeek and Sonnet for Act mode."
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Michael Lynch, professional developer
The Cline AI Assistant is MesmerizingPersonal blog post, mtlynch.io
Built a Zig tool that finds lexical illusions test-first with Cline: about 30 seconds and US$0.09 to get the first tests passing, and US$6 in OpenRouter credits for five hours of continuous work with file reads, writes and commands on auto-approve.
Practitioner account; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Positive
Paolo Galeone, independent developer
Using AI for Coding: My Journey with Cline and Large Language ModelsPersonal blog post, P. Galeone's blog
Used Cline in VS Code to redesign a Go SaaS side project; productivity rose sharply, but Sonnet's context limits forced task splitting and review stayed essential.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability
Counted for All practitioners.
Positive
Shivay Lamba, AI architect, Docker Captain and CNCF Ambassador
Early access to the Cline open-source Desktop app (X post)X post
On the free GLM-5.3 Flash model, Cline Desktop built a small Node.js task-list CLI. It passed all 15 tests after an initial environment issue was resolved and its handling of corrupted JSON was fixed.
"After resolving an initial environment issue and fixing how the CLI handled corrupted JSON, it successfully passed all 15 tests."
Practitioner account, pseudonymous; structured test; not counted, tied to the vendor
Positive
cdelsolar, Hacker News commenter
Comment on: an Ask HN thread on the moment generative AI first surprised peopleHacker News comment
Doubted coding agents until Cline built an entire working, non-trivial test app from one request.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
AUMOVIO and AWS, Martin Kraus and Amir Mahdi Namazi of AUMOVIO, with AWS authors
AUMOVIO Boosts Software Development with an Agentic Coding Assistant Powered by Amazon BedrockVendor case study, AWS for Industries blog
AUMOVIO's hackathon found Cline on Claude 3.7 Sonnet in Bedrock fixed a five-day bug in minutes, but general models duplicated internal libraries, so the pilot added fine-tuned models behind MCP.
Vendor material; ran the rollout; counts toward Delivered outcomes, Operational control
Positive
Rahul Ramesh, Consultant Developer at Thoughtworks
Research, Review, RebuildEngineering article, martinfowler.com
Ramesh migrated a Bahmni display control with Cline and Claude 3.5 Sonnet in under an hour for under US$2, with tests at about 95% coverage, against 3 to 6 days by hand. Cline read a style guide file each session, and line-by-line code review was the most significant challenge.
"Changing a single control would normally take 3 to 6 days of manual effort, but with these tools was completed in under an hour at a cost of under $2."
"This file is then read by Cline at the start of each session to ensure that all generated TypeScript code adheres to a consistent and recognized standard."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability
Positive
Dharanikumar Premkumar, Tech Lead and Application Architect at Thoughtworks
AI isn't just a coding partner, it can be a deployment partner, tooCompany blog post, Thoughtworks Insights
With little cloud experience, Premkumar used Cline with Claude to write the Terraform for his team's AWS platform and apply it through the AWS Terraform MCP server, reviewing every plan first. The generated README became the team's deployment playbook, though he had to cut oversized default resources.
"I again used the Cline and Claude combination to generate the Terraform configuration for its specific needs…"
"Before applying anything, I meticulously reviewed the terraform plan output."
Practitioner account, self-described; ran the rollout; counts toward Delivered outcomes, Operational control
Positive
Zichuan Xiong, Global Head of AIOps at Thoughtworks
AI evals for MCP in AIOpsCompany blog post, Thoughtworks Insights
Xiong and co-authors run Cline as the triage agent in an AIOps build for a LATAM bank, using Dynatrace MCP for log context. In their eval of an engineer's Cline sessions, one root cause analysis took three MCP tool calls and reported that logs and trace data were unavailable.
"…an AI agent (Cline) performs automatic triage and recommends context-aware resolutions by retrieving log context via Dynatrace MCP…"
"The agent successfully used three MCP tool calls to diagnose the issue."
Practitioner account, self-described; structured test; counts toward Delivered outcomes
Positive
Lorenz Hofmann-Wellenhof, engineer
How a Real AI Agent Works Under the HoodNewsletter post, The Growing Engineer (Substack)
After a couple of months on Cline, Hofmann-Wellenhof keeps Cursor for autocomplete and codebase questions but prefers Cline whenever an agent should work more freely on the codebase.
"I still use Cursor often, mainly for its autocomplete model (e.g., via the Tab key) and for asking questions about the codebase. But when it comes to letting an agent operate more freely on the codebase, I prefer Cline."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control
Positive
codethief, Hacker News commenter
Comment on: The unreasonable effectiveness of an LLM agent loop with tool useHacker News comment
Built an Android app prototype with Cline and Claude: several thousand lines with no compile errors; it debugged issues with feedback from codethief, though its fixes needed extra rounds.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Flux159, Hacker News commenter
Comment on: Ask HN: How much better are AI IDEs vs. copy pasting into chat apps?Hacker News comment
Moved to Cline with Gemini 2.5 Pro and a .clinerules file after Claude Code cost US$3 to US$5 per task. It runs commands and lint checks and costs little on a personal key.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
Jos van der Westhuizen, co-founder and CEO of Coplay
How Cline Quietly Changed The Game For Code CopilotsCompany blog post, Coplay blog
After a teammate's recommendation, Cline pushed through complex changes in Coplay's 60,000-line project where Cursor failed; single threads can pass US$6, which van der Westhuizen finds worth it.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Positive
inheritedwisdom, Hacker News commenter speaking for their team
Comment on: Ask HN: What's the best way to get started with LLM-assisted programing?Hacker News comment
Their team is rolling out VS Code, Cline and Sonnet 3.7 for data engineering, keeps an architecture README referenced on every task, and reports weeks taken off estimates.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Positive
Michael Lynch, professional developer
The Cline AI Assistant is MesmerizingPersonal blog post, mtlynch.io
Built a Zig tool that finds lexical illusions test-first with Cline: about 30 seconds and US$0.09 to get the first tests passing, and US$6 in OpenRouter credits for five hours of continuous work with file reads, writes and commands on auto-approve.
Practitioner account; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Positive
danielbln, Hacker News commenter, self-described
Comment on: Roo or Cline? We're building a supersetHacker News comment
Finds Cline better than Claude Code because it runs terminal commands, uses LSP and IDE diagnostics, and works with Gemini 2.5 alongside Anthropic models.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Thoughtworks Technology Advisory Board
Cline (Technology Radar, Trial)Technology Radar assessment, Thoughtworks Technology Radar
Places Cline in Trial as a strong supervised software engineering agent, citing Plan and Act, transparent token usage and MCP, while warning about token cost and rate limits.
Practitioner account; opinion; not counted until substantiated
Mixed
Birgitta Böckeler, Distinguished Engineer and AI-assisted delivery expert at Thoughtworks
The role of developer skills in agentic codingEngineering memo, martinfowler.com
Using agentic modes in Cursor, Windsurf and Cline on existing codebases, Böckeler steered constantly and often discarded changes, listing missteps that slow teams and hurt maintainability.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability
Counted for All practitioners.
Positive
rglover, Hacker News commenter, self-described
Comment on: Vibe coding and agentic engineering are getting closer than I'd likeHacker News comment
Uses Cline to approve or reject each turn and edit its changes before advancing, a practice rglover calls agentic engineering.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
rglover, Hacker News commenter
Comment on: Coding agents have replaced every framework I usedHacker News comment
Has used Cline with Sonnet and Opus daily for nearly two years, guiding each step; says this saves tokens and beats YOLO-mode results.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Vincent D. Warmerdam, data scientist and developer
cline feels like an upgradePersonal blog post, koaning.io
Plan mode made Cline click for Warmerdam, who values the token and cache transparency and checkpoint restore, with gripes about leftover terminals and missing shortcuts.
Practitioner account; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Positive
shmoogy, Hacker News commenter, self-described
Comment on: Ask HN: Cursor or Windsurf?Hacker News comment
Finds Cline works significantly better than Cursor on complex tasks because each request is paid and context is not truncated, and values seeing how it behaves.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Thoughtworks Technology Advisory Board
Cline (Technology Radar, Trial)Technology Radar assessment, Thoughtworks Technology Radar
Places Cline in Trial as a strong supervised software engineering agent, citing Plan and Act, transparent token usage and MCP, while warning about token cost and rate limits.
Practitioner account; opinion; not counted until substantiated
Positive
Shivay Lamba, AI architect, Docker Captain and CNCF Ambassador
Early access to the Cline open-source Desktop app (X post)X post
On the free GLM-5.3 Flash model, Cline Desktop built a small Node.js task-list CLI. It passed all 15 tests after an initial environment issue was resolved and its handling of corrupted JSON was fixed.
"After resolving an initial environment issue and fixing how the CLI handled corrupted JSON, it successfully passed all 15 tests."
Practitioner account, pseudonymous; structured test; not counted, tied to the vendor
Mixed
Juan Torchia, software architect
Cline on Autopilot: Why I Put Limits on My AgentPersonal blog post, DEV Community
Torchia tried Cline's auto-approve for everything and it worked, but on any codebase with history sets autonomy by action: full auto-approve for file reads, auto-approve with later diff review for edits in a bounded folder, manual approval for config edits, and destructive commands rejected by default.
Practitioner account, self-described; direct use; counts toward Operational control
Positive
NitpickLawyer, Hacker News commenter
Comment on: Why Cline doesn't index your codebaseHacker News comment
Likes the balance Cline strikes between autonomy and keeping the developer involved, and was surprised it worked well with local models such as Devstral.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
Jos van der Westhuizen, co-founder and CEO of Coplay
How Cline Quietly Changed The Game For Code CopilotsCompany blog post, Coplay blog
After a teammate's recommendation, Cline pushed through complex changes in Coplay's 60,000-line project where Cursor failed; single threads can pass US$6, which van der Westhuizen finds worth it.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Positive
Addy Osmani, engineer and author
Why I use Cline for AI EngineeringNewsletter, Elevate (addyo.substack.com)
After extensively testing Cursor, WindSurf and GitHub Copilot, Osmani finds Cline uniquely valuable for model flexibility, Plan/Act, checkpoints, MCP extensibility and cost accounting during a session, while noting token-based pricing requires attention to model selection.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Counted for All practitioners.
Positive, 9
Positive
Marha01, Hacker News commenter
Comment on: Fable 5 is BackHacker News comment
Uses Cline's automatic model switch between Plan and Act: Opus for planning, Sonnet for implementation.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
chisleu, Hacker News commenter
Comment on: Claude Opus 4.1Hacker News comment
Uses Opus or Gemini 2.5 Pro for Plan mode and Sonnet for Act mode in Cline, finding Opus better at architectural problems.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
Vincent D. Warmerdam, data scientist and developer
cline feels like an upgradePersonal blog post, koaning.io
Plan mode made Cline click for Warmerdam, who values the token and cache transparency and checkpoint restore, with gripes about leftover terminals and missing shortcuts.
Practitioner account; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Positive
chrisnolet, Hacker News commenter
Comment on: Show HN: Vibe coding from your phoneHacker News comment
Calls Cline the dark horse of AI editors and argues engineers should spend about four times as long in Plan mode as in Act mode, reviewing the plan first.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
theshrike79, Hacker News commenter
Comment on: Perverse incentives of vibe codingHacker News comment
Settled on Cline through OpenRouter, with Gemini 2.5 for Plan mode and Claude 3.7 for Act mode, after finding Claude Code and Aider overreached.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
Addy Osmani, engineer and author
What's new in @cline? (X post)X post
Osmani plans in Cline with Gemini or DeepSeek and executes with Sonnet in Act mode, a split he uses to keep token costs down, and recommends Cline to Cursor and Windsurf users.
"I use Plan mode with Gemini or DeepSeek and Sonnet for Act mode."
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Addy Osmani, engineer and author
Why I use Cline for AI EngineeringNewsletter, Elevate (addyo.substack.com)
After extensively testing Cursor, WindSurf and GitHub Copilot, Osmani finds Cline uniquely valuable for model flexibility, Plan/Act, checkpoints, MCP extensibility and cost accounting during a session, while noting token-based pricing requires attention to model selection.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Lorenz Hofmann-Wellenhof, engineer
How a Real AI Agent Works Under the HoodNewsletter post, The Growing Engineer (Substack)
After a couple of months on Cline, Hofmann-Wellenhof keeps Cursor for autocomplete and codebase questions but prefers Cline whenever an agent should work more freely on the codebase.
"I still use Cursor often, mainly for its autocomplete model (e.g., via the Tab key) and for asking questions about the codebase. But when it comes to letting an agent operate more freely on the codebase, I prefer Cline."
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control
Positive
Thoughtworks Technology Advisory Board
Cline (Technology Radar, Trial)Technology Radar assessment, Thoughtworks Technology Radar
Places Cline in Trial as a strong supervised software engineering agent, citing Plan and Act, transparent token usage and MCP, while warning about token cost and rate limits.
Practitioner account; opinion; not counted until substantiated
Mixed or negative, 1
Mixed
umair1808, Cline user filing a GitHub issue
Plan mode not respecting edit permissions + degraded task quality after v4.x SDK migrationGitHub issue
After the 4.x SDK migration, Cline edits files while still in Plan mode and task quality dropped, on DeepSeek through a self-hosted LiteLLM.
Practitioner account, pseudonymous; direct use; counts toward Reliability, Operational control
Every concern practitioners raise, most raised first.
Counted for All practitioners.
Raised as a concern, 17
Mixed
nicolaspe, GitHub user
Broken Anthropic prompt caching on tool calls in v4.1.25 ("Next" engine) resulting in cost spikes.GitHub issue on cline/cline
After moving to Cline 4.1.25's Next engine, prompt caching stopped advancing during tool loops, so a 70,000-token task paid the full input rate on every step; reverting to 4.0.12 restored cheap loops.
"Because of this, a 70k context task bills the full $2.00/M input rate on every single autonomous step instead of the $0.10/M cached rate."
Practitioner account, pseudonymous; structured test; counts toward Reliability, Cost predictability and limits
Mixed
romemozu, ClinePass and Cline Desktop user
Counted 1,156 calls across 10 Desktop sessions: failed sessions kept running for up to 23 hours, one used 112 million input tokens, and models labeled free drew the credit balance negative.
Practitioner account, pseudonymous; direct use; counts toward Reliability, Cost predictability and limits
Mixed
TrueBlueKiwi, Hacker News commenter
Comment on: Ask HN: What are tools you have made for yourself since the advent of AI?Hacker News comment
Building a web app with Cline and Claude Sonnet 4.7, TrueBlueKiwi finds it re-reads the project documentation on every turn, which costs tokens.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Negative
chankstein38, Hacker News commenter
Comment on: AI's economics don't make senseHacker News comment
Put US$15 into an API account for Cline and ran out within a few hours on a simple project; estimates US$750 a month at that rate.
Practitioner account, pseudonymous; direct use; counts toward Cost predictability and limits
Mixed
tgtweak, Hacker News commenter
Comment on: GPT-5.2Hacker News comment
Running Claude with a 1M-token context window on a long project, mostly in Cline, gets expensive.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Mixed
andreas83, Cline user filing a GitHub issue
Many Unexpected API Response still burning creditsGitHub issue
With Grok 3 and Gemini 2.5 Pro, repeated unexpected API responses and tool-format retries used 80% of the account's credits.
Practitioner account, pseudonymous; structured test; counts toward Reliability, Cost predictability and limits
Negative
qinsig, Hacker News commenter, self-described
Comment on: Generative AI coding tools and agents do not work for meHacker News comment
Advises avoiding agents that blow through money, such as Cline or Claude Code on an API key, in favor of a Claude Pro subscription.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Negative
bicepjai, Hacker News commenter
Comment on: A flat pricing subscription for Claude CodeHacker News comment
Burned US$200 in a weekend on a Rust side project with Cline and Claude, then US$30 in 30 minutes on Gemini 2.5 Pro, and moved to DeepSeek and local models.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
int_19h, Hacker News commenter, self-described
Comment on: Ask HN: Cursor or Windsurf?Hacker News comment
Says Cline "seems to be the best thing", suspecting it does no context trimming under the hood to keep costs down. For the same reason, watching its token and dollar counter tick as it works is not exactly fun.
not counted, repeats a stronger account
Positive
asaddhamani, Hacker News commenter
Comment on: A flat pricing subscription for Claude CodeHacker News comment
Says Cline gets better results than Cursor or Copilot because it does not trim context, though asaddhamani has spent US$20 of API usage in a single day.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
SkyPuncher, Hacker News commenter
Comment on: Claude's Max PlanHacker News comment
Can easily burn US$20 an hour of credits in Cline or Roo Code, and full-time development could pass US$100 a week; SkyPuncher switched to Cursor.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Thoughtworks Technology Advisory Board
Cline (Technology Radar, Trial)Technology Radar assessment, Thoughtworks Technology Radar
Places Cline in Trial as a strong supervised software engineering agent, citing Plan and Act, transparent token usage and MCP, while warning about token cost and rate limits.
Practitioner account; opinion; not counted until substantiated
Mixed
timbritt, Hacker News commenter
Comment on: Ask HN: What's the best way to get started with LLM-assisted programing?Hacker News comment
Building with Cline and Roo, timbritt warns the tools waste time and money without a solid approach, and recommends a memory bank and strict test-first workflow to contain looping spend.
Practitioner account, pseudonymous; direct use; counts toward Reliability
Positive
kojix2, developer
Building My First Web App with the Help of AIPersonal blog post, DEV Community
Built and deployed a first web app with VS Code, Cline and Claude 3.7 Sonnet in about 10 hours; the AI itself cost close to US$70.
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Mixed
t1amat, Hacker News commenter who provisions tools for a team
Comment on: GPT-5 for DevelopersHacker News comment
Provisioned the team with Claude Code subscriptions after calling API pay-per-use prohibitively costly; notes Cline's Claude Code provider but finds Claude Code as good or better.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Mixed
theturtletalks, Hacker News commenter
Comment on: Kiro: A new agentic IDEHacker News comment
Used Cline and Roo with Sonnet through OpenRouter, then moved to Claude Code on the Max plan, which costs less than API credits and gives better results.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
jjmarr, Hacker News commenter
Comment on: Claude 4Hacker News comment
A junior developer shipped an LLVM change after Cline found the 5 places to edit across 10,000+ files, and notes Claude costs hundreds of dollars a month.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Discussed without concern, 12
Positive
Jos van der Westhuizen, co-founder and CEO of Coplay
How Cline Quietly Changed The Game For Code CopilotsCompany blog post, Coplay blog
After a teammate's recommendation, Cline pushed through complex changes in Coplay's 60,000-line project where Cursor failed; single threads can pass US$6, which van der Westhuizen finds worth it.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Positive
rglover, Hacker News commenter
Comment on: Coding agents have replaced every framework I usedHacker News comment
Has used Cline with Sonnet and Opus daily for nearly two years, guiding each step; says this saves tokens and beats YOLO-mode results.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
esperent, Hacker News commenter
Comment on: Claude Code 2.0Hacker News comment
Found Claude Code very expensive and no better than Cline on similar tasks, so esperent stays with Cline and switches APIs by price and performance.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
ramesh31, Hacker News commenter
Comment on: AWS CEO says using AI to replace junior staff is 'Dumbest thing I've ever heard'Hacker News comment
Used Cline heavily at work with over US$1,000 a month in token spend before switching to Claude Code, and says the underlying Claude model matters most.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
KeysToHeaven, Hacker News commenter
Comment on: a quote-reply extension for Slack threads, built with GPT-5Hacker News comment
Built and published a Chrome extension with Cline and GPT-5 in about 30 minutes for under a dollar.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Positive
Eleanor Berger, developer
LISP Interpreter: Created by Cline and Qwen 3 Coder for less than US$1Project write-up, GitHub Gist (intellectronica)
Built a working Scheme-like Lisp interpreter in Rust with Cline and Qwen 3 Coder in 10 minutes for under one US dollar.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Positive
shmoogy, Hacker News commenter, self-described
Comment on: Ask HN: Cursor or Windsurf?Hacker News comment
Finds Cline works significantly better than Cursor on complex tasks because each request is paid and context is not truncated, and values seeing how it behaves.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Flux159, Hacker News commenter
Comment on: Ask HN: How much better are AI IDEs vs. copy pasting into chat apps?Hacker News comment
Moved to Cline with Gemini 2.5 Pro and a .clinerules file after Claude Code cost US$3 to US$5 per task. It runs commands and lint checks and costs little on a personal key.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
razemio, Hacker News commenter
Comment on: OpenAI Codex CLI: Lightweight coding agent that runs in your terminalHacker News comment
Rates Cline above Aider; Gemini 2.5 Pro is cheaper in Cline but often fails to use its tools, so razemio pays more for Sonnet to save time.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Addy Osmani, engineer and author
What's new in @cline? (X post)X post
Osmani plans in Cline with Gemini or DeepSeek and executes with Sonnet in Act mode, a split he uses to keep token costs down, and recommends Cline to Cursor and Windsurf users.
"I use Plan mode with Gemini or DeepSeek and Sonnet for Act mode."
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Michael Lynch, professional developer
The Cline AI Assistant is MesmerizingPersonal blog post, mtlynch.io
Built a Zig tool that finds lexical illusions test-first with Cline: about 30 seconds and US$0.09 to get the first tests passing, and US$6 in OpenRouter credits for five hours of continuous work with file reads, writes and commands on auto-approve.
Practitioner account; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Positive
Addy Osmani, engineer and author
Why I use Cline for AI EngineeringNewsletter, Elevate (addyo.substack.com)
After extensively testing Cursor, WindSurf and GitHub Copilot, Osmani finds Cline uniquely valuable for model flexibility, Plan/Act, checkpoints, MCP extensibility and cost accounting during a session, while noting token-based pricing requires attention to model selection.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.Cline itself is free and shows token use and cost per task. Spend limits sit with the provider account, Cline credits, or ClinePass quotas. A report of high spend does not establish that spend cannot be capped. Authorization
Open them in the recordCounted for All practitioners.
Raised as a concern, 12
Mixed
nicolaspe, GitHub user
Broken Anthropic prompt caching on tool calls in v4.1.25 ("Next" engine) resulting in cost spikes.GitHub issue on cline/cline
After moving to Cline 4.1.25's Next engine, prompt caching stopped advancing during tool loops, so a 70,000-token task paid the full input rate on every step; reverting to 4.0.12 restored cheap loops.
"Because of this, a 70k context task bills the full $2.00/M input rate on every single autonomous step instead of the $0.10/M cached rate."
Practitioner account, pseudonymous; structured test; counts toward Reliability, Cost predictability and limits
Negative
Neywiny, Hacker News commenter, self-described
Comment on: Which tools do Claude, Codex and Cursor choose? We measured 17k runs to find outHacker News comment
Sees no improvement since Cline's new harness reached full rollout in August 2026; agents abandon its tools after one try and fall back to raw commands.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
umair1808, Cline user filing a GitHub issue
Plan mode not respecting edit permissions + degraded task quality after v4.x SDK migrationGitHub issue
After the 4.x SDK migration, Cline edits files while still in Plan mode and task quality dropped, on DeepSeek through a self-hosted LiteLLM.
Practitioner account, pseudonymous; direct use; counts toward Reliability, Operational control
Negative
Stanislav Nabatnikov, developer
Why the Cline Extension for VS Code Fails as an "Autonomous Agent" Under LinuxPersonal blog post, DEV Community
On a real Linux project with auto-approve on, Cline froze twice on its first task when shell escaping and a multi-line commit message broke its terminal handling.
Practitioner account, self-described; direct use; counts toward Reliability
Mixed
stefanCCS, Cline user filing a GitHub issue, self-described
Cline hangs completely after skipping a command execution, requires VS Code restartGitHub issue
Skipping a proposed command repeatedly leaves Cline unresponsive until VS Code restarts, even after adding .clinerules for terminal commands.
Practitioner account, pseudonymous; direct use; counts toward Reliability
Negative
Robat Williams, developer building client web applications at Scott Logic
GLM-5.2: Considerations for enterprise teams starting out with open-weight modelsCompany blog post, Scott Logic Blog
Setting up an open-weight model trial, Scott Logic's team shortlisted three coding agents and chose OpenCode. Cline was dropped after the team hit bugs and missing features while trialling it.
"Cline had a couple of issues/omissions and bugs we ran into while trialling."
"We shortlisted some well known ones to briefly trial: Pi, OpenCode and Cline. OpenCode was our agent of choice."
Practitioner account, self-described; direct use; not counted, takes no position
Negative
lewisl, Cline user filing a GitHub issue
Cline 4.0.0 issue: MCP server settings no longer read after upgradingGitHub issue
Spent over an hour failing to connect an MCP server on a fresh Cline 4.0.0 install because settings went to a file it no longer read. Then lewisl reverted to 3.89.2 and wrote that the release showed the team could not be trusted.
Practitioner account, pseudonymous; direct use; counts toward Reliability, Adoption and upkeep
Negative
jszymborski, Hacker News commenter
Comment on: A guide to local coding modelsHacker News comment
Running local models, jszymborski traced repeated failures to Cline itself, which mishandles empty files in those projects.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Negative
Revenoti, Cline user filing a GitHub issue, self-described
Saw files Cline was not working on appear as deleted, repeatedly, and stopped using Cline until it is fixed.
Practitioner account, pseudonymous; direct use; counts toward Reliability
Negative
lizardking, Hacker News commenter, self-described
Comment on: Cognition (Devin AI) to Acquire WindsurfHacker News comment
Continue and Cline bugged out so badly and so often that lizardking did not find them worth using.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
romemozu, ClinePass and Cline Desktop user
Counted 1,156 calls across 10 Desktop sessions: failed sessions kept running for up to 23 hours, one used 112 million input tokens, and models labeled free drew the credit balance negative.
Practitioner account, pseudonymous; direct use; counts toward Reliability, Cost predictability and limits
Mixed
rvnlord, Cline user filing a GitHub issue
Cline truncates tool output & file content in context (forces splitting large files into chunks)GitHub issue
Cline cuts command output at 8,000 characters and file reads at 50,000, forcing extra round trips and chunked edits; rvnlord asks for configurable limits.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Reliability, Adoption and upkeep
Discussed without concern, 3
Mixed
_fat_santa, Hacker News commenter
Comment on: ScreenCoder: An intelligent UI-to-code generation systemHacker News comment
Suggests a day of building UI components with a Figma MCP server and adding rules as the AI errs; the first rule told Cline to use the MCP server.
Practitioner account, pseudonymous; ran the rollout; not counted, repeats a stronger account
Positive
Vincent D. Warmerdam, data scientist and developer
cline feels like an upgradePersonal blog post, koaning.io
Plan mode made Cline click for Warmerdam, who values the token and cache transparency and checkpoint restore, with gripes about leftover terminals and missing shortcuts.
Practitioner account; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits
Positive
efitz, Hacker News commenter, self-described
Comment on: OpenAI's Windsurf deal is off, and Windsurf's CEO is going to GoogleHacker News comment
Calls Cline and Roo Code fantastic as normal VS Code extensions, notes they occasionally lose the terminal connection, and prefers bringing a personal key over an IDE middleman.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Counted for All practitioners.
Mixed
LittleRainC, GitHub user
[Bug]: Local Ollama models (Qwen 2.5 Coder) trapped in infinite loop emitting raw JSON instead of triggering tools due to strict XML parser constraintGitHub issue on cline/cline
With qwen2.5-coder:32b on Ollama, Cline ignored the model's JSON tool calls as plain text, so the model looped on the same call until a custom rule forced Cline's XML format.
"However, Cline's current streaming parser completely ignores this raw JSON output and treats it as plain conversational text."
Practitioner account, pseudonymous; structured test; counts toward Reliability
Negative
stuaxo, Hacker News commenter
Comment on: Local AI needs to be the normHacker News comment
Tried Cline and couldn't get it working well, partly because at the time it expected OpenAI's output format.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
wickedsight, Hacker News commenter
Comment on: AI coding assistants are getting worse?Hacker News comment
Uses Cline with a local Qwen 3 32B at work, where cloud AI is not allowed; basic requests loop without useful output, while cloud Codex handles complex tasks.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Negative
John-Codes, Cline user filing a GitHub issue
"Cline uses complex prompts and iterative task execution that may be challenging for less capable models..."GitHub issue
Running Qwen3 Coder on Cline v3.26.7, asks Cline to remove its warning that recommends Claude 4 Sonnet for less capable models, saying it made open-source LLMs impossible to use. Says they will switch tools soon if it continues.
Practitioner account, pseudonymous; direct use; counts toward Reliability, Operational control
Negative
NicoJuicy, Hacker News commenter
Comment on: Open models by OpenAIHacker News comment
Found gpt-oss 20B on a local RTX 3090 unusable in Cline: tool calls failed and the model did not wait for feedback.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
razemio, Hacker News commenter
Comment on: OpenAI Codex CLI: Lightweight coding agent that runs in your terminalHacker News comment
Rates Cline above Aider; Gemini 2.5 Pro is cheaper in Cline but often fails to use its tools, so razemio pays more for Sonnet to save time.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
AUMOVIO and AWS, Martin Kraus and Amir Mahdi Namazi of AUMOVIO, with AWS authors
AUMOVIO Boosts Software Development with an Agentic Coding Assistant Powered by Amazon BedrockVendor case study, AWS for Industries blog
AUMOVIO's hackathon found Cline on Claude 3.7 Sonnet in Bedrock fixed a five-day bug in minutes, but general models duplicated internal libraries, so the pilot added fine-tuned models behind MCP.
Vendor material; ran the rollout; counts toward Delivered outcomes, Operational control
Mixed
ramesh31, Hacker News commenter
Comment on: I want everything local - Building my offline AI workspaceHacker News comment
Tried current Qwen and GLM models through OpenRouter in Cline and found them roughly on par with Claude 3.0, well short of Claude for coding.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Mixed
andreas83, Cline user filing a GitHub issue
Many Unexpected API Response still burning creditsGitHub issue
With Grok 3 and Gemini 2.5 Pro, repeated unexpected API responses and tool-format retries used 80% of the account's credits.
Practitioner account, pseudonymous; structured test; counts toward Reliability, Cost predictability and limits
Mixed
neilsharma, Hacker News commenter
Comment on: Ask HN: How much better are AI IDEs vs. copy pasting into chat apps?Hacker News comment
Uses Cline with .clinerules for React work, narrows tasks so Gemini Flash or Haiku can handle them, saw no gain from paid models, and finds Gemini 2.5 Pro over-reaches.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Counted for All practitioners.
Mixed
IRwDONE, GitHub user
Agent silently set CLINE_CAPTURE_PROVIDER_REQUEST=full → Plaintext credentials on disk → Unauthorized paid calls on an unrelated provider keyGitHub issue on cline/cline
With tool auto-approval on by default in Cline Desktop, an agent set request capture to full without asking, writing credentials to disk in 75 of 320 captures. A later run used a key the user never shared for 11 paid requests.
"A later agent run then read an API key that the user had never shared and used it to make 11 paid requests on the user's own provider account, without asking and without mentioning cost."
Practitioner account, pseudonymous; direct use; counts toward Operational control
Mixed
Korn94, Cline user on Windows
Reports that in Act mode on Windows, a cleanup command run through Cline's run_commands tool had mangled PowerShell quoting, so rmdir ran against the data drive root and deleted about 1.5 TB of photos, videos and documents. Only code under git and pushed was recovered.
Practitioner account, pseudonymous; direct use; counts toward Reliability, Operational control
Mixed
Juan Torchia, software architect
Cline on Autopilot: Why I Put Limits on My AgentPersonal blog post, DEV Community
Torchia tried Cline's auto-approve for everything and it worked, but on any codebase with history sets autonomy by action: full auto-approve for file reads, auto-approve with later diff review for edits in a bounded folder, manual approval for config edits, and destructive commands rejected by default.
Practitioner account, self-described; direct use; counts toward Operational control
Mixed
BroadlyWhitaker, Cline user filing a GitHub issue, self-described
Git commit --amend should not be considered a "safe command"GitHub issue
New to Cline, BroadlyWhitaker found that with safe commands auto-approved it amends git commits without asking, which the report calls destructive.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Negative
wunderwuzzi, security researcher
Cline: Vulnerable To Data Exfiltration And How To Protect Your DataSecurity research post, Embrace The Red
Showed a prompt injection that makes Cline read a .env file and leak it through a rendered image, recommending auto-approve off by default; unfixed 90+ days after disclosure.
Practitioner account, self-described; structured test; counts toward Operational control, Adoption and upkeep
Negative
Revenoti, Cline user filing a GitHub issue, self-described
Saw files Cline was not working on appear as deleted, repeatedly, and stopped using Cline until it is fixed.
Practitioner account, pseudonymous; direct use; counts toward Reliability
Mixed
getsec, for a team building an MCP server on internal APIs
With only safe commands auto-approved, Cline ran an eval of a base64-encoded command hidden in content it fetched from an API. The getsec team used this to show it could be tricked into running arbitrary code.
Practitioner account, pseudonymous; structured test; counts toward Operational control
Mixed
chezsmithy, for development teams in a regulated company
MCP MarketPlace Enterprise UseGitHub issue
Says their highly regulated company vets every MCP server through infosec before developers get it from an internal source, and Cline 3.4.0's MCP marketplace makes pulling unapproved code too easy. Asks for a way to disable the marketplace or redirect it to an internal registry.
Practitioner account, pseudonymous; ran the rollout; counts toward Operational control
Mixed
umair1808, Cline user filing a GitHub issue
Plan mode not respecting edit permissions + degraded task quality after v4.x SDK migrationGitHub issue
After the 4.x SDK migration, Cline edits files while still in Plan mode and task quality dropped, on DeepSeek through a self-hosted LiteLLM.
Practitioner account, pseudonymous; direct use; counts toward Reliability, Operational control
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.In the IDE extensions, edits and commands wait for approval until a team turns on Auto Approve. The CLI auto-approves by default, and no Cline product yet enforces the organization YOLO-mode setting. Enterprise provider configuration
Open them in the recordCounted for All practitioners.
Raised as a concern, 6
Mixed
rvnlord, Cline user filing a GitHub issue
Cline truncates tool output & file content in context (forces splitting large files into chunks)GitHub issue
Cline cuts command output at 8,000 characters and file reads at 50,000, forcing extra round trips and chunked edits; rvnlord asks for configurable limits.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Reliability, Adoption and upkeep
Mixed
khkiley, Cline user filing a GitHub issue
Agressive context compaction with opus-4.6:1mGitHub issue
With Opus 4.6 and a 1M window, Cline compacted at under 200,000 tokens mid-task, and khkiley finds it gets a little lost afterward.
Practitioner account, pseudonymous; direct use; counts toward Reliability
Mixed
mosmondor, Cline user filing a GitHub issue
Automatic chat compacting at 160k tokens loses critical context and causes Cline to restart completed workGitHub issue
After automatic compaction above 160,000 tokens, Cline loses track of finished work and starts redoing it; mosmondor asks for control over compaction.
Practitioner account, pseudonymous; direct use; counts toward Reliability
Positive
Paolo Galeone, independent developer
Using AI for Coding: My Journey with Cline and Large Language ModelsPersonal blog post, P. Galeone's blog
Used Cline in VS Code to redesign a Go SaaS side project; productivity rose sharply, but Sonnet's context limits forced task splitting and review stayed essential.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability
Mixed
TrueBlueKiwi, Hacker News commenter
Comment on: Ask HN: What are tools you have made for yourself since the advent of AI?Hacker News comment
Building a web app with Cline and Claude Sonnet 4.7, TrueBlueKiwi finds it re-reads the project documentation on every turn, which costs tokens.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
chaboud, Hacker News commenter
Comment on: OpenAI o3 and o4-miniHacker News comment
Gets great results from Cline with Sonnet 3.7 on AWS Bedrock for smaller projects in popular languages, but warns purely prompt-built projects degrade as context windows saturate.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Discussed without concern, 2
Positive
a_e_k, Hacker News commenter
Comment on: 1M context is now generally available for Opus 4.6 and Sonnet 4.6Hacker News comment
Adopting Cline at work on an enterprise model plan with a 1M-token window, a_e_k finds it holds up to about 700,000 tokens in hands-on pair-programming mode.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
tgtweak, Hacker News commenter
Comment on: GPT-5.2Hacker News comment
Running Claude with a 1M-token context window on a long project, mostly in Cline, gets expensive.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Statements this group's sources agree on, then the ones close to agreement.
Close to agreement: each needs 19 supporting sources
Cline gives the best results on small, well-scoped tasks with tests or clear checks, and drifts on large open-ended changes.
7 supporting sourcesOn self-directed work, Cline shortens the path from an idea to a working change, at a token cost the builder can see.
7 supporting sourcesCline shows token use and cost as a task runs, which helps decide which model gets which work.
6 supporting sourcesThe model a team configures, more than Cline itself, decides the quality of the result.
6 supporting sourcesRules files, a memory bank, or an architecture README referenced on every task make Cline's results more consistent.
6 supporting sourcesQuestions the sources split on, with the count on each side.
Evidence remains mixed
Pay per token, 10 sources
Some prefer paying the provider directly, with any model and no middleman, because costs stay low for bursty or careful use.
Flat subscription, 6 sources
Others moved to a subscription product such as Cursor or Claude Code because per-token spend on heavy use ran too high.
Evidence remains mixed
Cheaper models work, 11 sources
Some report good results from DeepSeek, Qwen, Devstral, or Gemini Flash at a fraction of frontier prices.
Frontier models needed, 3 sources
Others found cheaper or local models looped or failed Cline's tool calls, and fell well short of Claude.
Evidence remains mixed
Approve each step, 10 sources
Some keep approvals on and steer turn by turn, rejecting or editing changes before Cline moves on.
Auto-approve and review at the end, 3 sources
Others auto-approve edits and commands and let Cline iterate until tests pass, then review the result.
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 All practitioners. Leaders and teams have 8 accounts; GILD prints their sentiment from 20.
14 sources. These accounts focus on spend control on your own keys, approval policy, governance, and what changes when Cline moves from individual use to a team.
80 sources. These accounts focus on model and provider choice, spend on personal keys, speed, and how Cline performs in self-directed work.
94 sources. This count includes every practitioner: individual builders and the few team and leadership accounts together.
94 sources from practitioners, by the context they work in. Filled rows count for this group.
Emphasizes Model choice and spend on personal keys, then reliability after updates
Emphasizes Command safety and MCP vetting, with shared rules files
Emphasizes Pilots on company model accounts, and model limits on internal code
Emphasizes Supervision and token cost in a trial
What changes with scale. Individual builders emphasize model choice and what their own keys cost. Team and leadership accounts turn to supervision. They ask which commands and MCP servers the agent may run, and whether per-token spend beats a flat-rate seat across the team.
Where this group's sources were published.
Counted for All practitioners, from 94 sources. Change group
112 sources for this group, 18 of them documentation
Older evidence is retained when it still applies. Sources that depend on a changed release, a fixed bug, or a closed issue 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 112 sourcesAbout this read
A running review of what engineering teams report about Cline, from individual builders to engineering leaders. 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 0 of 8, individual builders 4 of 65, all practitioners 4 of 73.
Each row opens its sources for the group you are reading.
| Finding | Engineering leaders and teams | Individual builders | All practitioners | Sources |
|---|---|---|---|---|
| Current readCurrent sentiment | 5 positive, 2 mixed, 1 negative | 38 positive, 15 mixed, 12 negative | 43 positive, 17 mixed, 13 negative | Sources |
| PraiseModel and Provider Choice | 0 of 0 positive (under the minimum) | 20 of 20 positive | 20 of 20 positive | Sources |
| PraisePlan and Act | 1 of 1 positive (under the minimum) | 8 of 9 positive | 9 of 10 positive | Sources |
| PraiseVisibility and Control | 2 of 2 positive (under the minimum) | 8 of 8 positive | 10 of 10 positive | Sources |
| PraiseMulti-Step Execution | 8 of 8 positive | 7 of 7 positive | 15 of 15 positive | Sources |
| ConcernToken Spend on Your Own Key | 2 of 3 raise it (under the minimum) | 15 of 26 raise it | 17 of 29 raise it | Sources |
| ConcernContext Management | 0 of 0 raise it (under the minimum) | 6 of 8 raise it | 6 of 8 raise it | Sources |
| ConcernReliability and Update Regressions | 1 of 1 raise it (under the minimum) | 11 of 14 raise it | 12 of 15 raise it | Sources |
| ConcernResults Depend on the Model | 1 of 1 raise it (under the minimum) | 9 of 9 raise it | 10 of 10 raise it | Sources |
| ConcernApprovals and Safety | 2 of 2 raise it (under the minimum) | 7 of 7 raise it | 9 of 9 raise it | Sources |
| ConsensusStrongest on scoped work | 0 supporting (under the minimum) | 7 supporting (below threshold) | 7 supporting (below threshold) | Sources |
| ConsensusFaster on self-directed work | 0 supporting (under the minimum) | 7 supporting (below threshold) | 7 supporting (below threshold) | Sources |
| ConsensusSpend is visible per task | 1 supporting (under the minimum) | 5 supporting (below threshold) | 6 supporting (below threshold) | Sources |
| ConsensusThe model decides the result | 1 supporting (under the minimum) | 5 supporting (below threshold) | 6 supporting (below threshold) | Sources |
| ConsensusWritten rules make results consistent | 2 supporting (under the minimum) | 4 supporting (under the minimum) | 6 supporting (below threshold) | Sources |
| DisagreementIs paying per token on your own key better value than a flat subscription? | Pay per token 2, Flat subscription 1 (under the minimum) | Pay per token 8, Flat subscription 5 | Pay per token 10, Flat subscription 6 | Sources |
| DisagreementDo cheaper, open-weight, or local models do the job in Cline? | Cheaper models work 0, Frontier models needed 0 (under the minimum) | Cheaper models work 11, Frontier models needed 3 | Cheaper models work 11, Frontier models needed 3 | Sources |
| DisagreementHow much should Cline do before a human approves the next step? | Approve each step 3, Auto-approve and review at the end 0 (under the minimum) | Approve each step 7, Auto-approve and review at the end 3 | Approve each step 10, Auto-approve and review at the end 3 | Sources |
| Current readEvidence mix | 32 sources | 98 sources | 112 sources | Sources |
GILD's checks of claims against official documentation, kept apart from the source counts.
