Claude Code vs Devin for Recurring Repository Work

7 min read Comparisons

Compare Claude Code and Devin on recurring coding work, memory boundaries, verification evidence, execution options and pricing.

Claude Code and Devin both run recurring coding work, but where will yesterday's correction live when the next automated session starts? Claude Code's memory guide covers editable local notes, while Devin's memory guide explicitly excludes automation sessions from personal memory. A team choosing an agent for repeat repository maintenance needs to settle that guidance path before trusting the schedule.

The issue is narrower than whether either product remembers anything. Interactive corrections, project instructions and unattended runs have different boundaries. Compare a recurring test-maintenance job by tracing the exact instruction it should reuse, the evidence it must produce and the person who reviews the result.

Claude Code coordinates sessions and threads

Claude Code supports local terminal and editor work as well as cloud sessions. Its Projects documentation describes a coordinator that starts and tracks threads, with editable project memory read by later cloud threads. Projects are public beta, rolling out gradually on Pro, Max and Ultra, and are not yet offered on Team or Enterprise.

Experimental agent teams add a lead that assigns work and synthesizes results, with agent-owned tasks, dependencies and completion states. The documentation does not describe the same work item being owned interchangeably by a person and an agent. A team should inspect that ownership model separately from the coordinator's ability to dispatch parallel work.

For repeat processes, Dynamic workflows place loops, branches and agent stages in a JavaScript script. The script controls orchestration and can be rerun. This is a published execution model, so the comparison should not reduce Claude Code to an open terminal waiting for a prompt.

Devin connects delegated work to delivery

Devin coordinates engineering requests through repository-connected sessions. Its Linear integration guide describes assigning a ticket to Devin, following its activity feed and synchronized plan, and sending follow-up messages through the ticket's agent session. The work record stays in Linear.

A coordinator session can scope a larger job, delegate to managed sessions, monitor progress and resolve conflicts. That coordinates a delegated project, not a permanent team lead. Ask Devin also helps scope an implementation before passing a context-rich prompt into an Agent session.

Its Dynamic Workflows guide describes deterministic Python scripts that order agents and transfer structured results. Calls are recorded. After interruption, completed results replay while unfinished work runs again, giving the team a concrete recovery behavior to test.

Which corrections reach recurring work?

Claude Code's memory guide describes auto memory that retains local corrections and preferences in files you can read, edit or delete. Public-beta Projects have separate editable memory for later cloud threads. Those are published memory scopes, not evidence that every saved instruction reaches every scheduled run.

Its research-preview scheduled tasks run from schedules, HTTP calls or GitHub events. Each occurrence opens a cloud session, with history accessible through the saved task. The tasks belong to an individual account, are not shared with teammates and act through the owner's connected identities.

Devin's personal memory guide says preferences and corrections carry into later sessions and can be browsed, changed or removed. It also explicitly says automation sessions do not read or write that memory. A correction saved during interactive work therefore needs another documented route into unattended automation.

Both products have recurring execution, but a schedule does not establish shared ownership or durable guidance. For your maintenance job, revise an instruction and examine what the next unattended session actually receives. In Claude Code, distinguish local memory, project memory and scheduled-task configuration. In Devin, do not use a personal-memory correction as evidence that an Automation learned the instruction.

What evidence makes a change ready?

Claude Code's /goal guide defines a measurable completion condition checked by a separate model after each turn. When the condition is unmet, the work continues. The evaluator judges conversation evidence and does not independently run commands or read files, so a test result needs trustworthy evidence in that conversation.

Devin's testing guide describes building a focused test plan from a code change, executing it in the browser and sending a recording. Its beta custom reviewers can also check pull requests against written instructions with configurable passing conditions. Browser evidence and a turn-by-turn completion evaluator answer different review questions.

Claude Code also publishes hosted Code Review on Team and Enterprise plans against CLAUDE.md and REVIEW.md rules. Research-preview ultrareview sends separate reviewers into a cloud sandbox and independently reproduces and verifies findings. That is a concrete independent-review path, not a capability the team must obtain by leaving Claude Code.

Devin's Review documentation describes repository instructions and beta custom reviewers for focused concerns. It does not establish those reviewers as independent second reviews specifically triggered by risky changes. Beta Autoapprover assesses intent, quality and incident risk, but publishes an approval decision with reasons, not an overall merge-risk rating.

Neither product should be accepted on a session status alone. Claude Code explicitly warns that Succeeded on a scheduled run does not establish completion of the prompt. For either candidate, require the regression test, reviewer finding or browser recording that actually addresses the maintenance job's acceptance condition.

Where execution and billing stop

Claude Code runs locally and in cloud sessions, and Projects can hand a thread to your computer through Remote Control. Local permission rules can allow, ask or deny tool actions. Cloud project threads do not apply repository permission rules, so review that boundary before assuming the local approval policy carries over.

Devin offers cloud sessions and Outposts for commands, file edits and repository access on your own machines. Planning and inference remain in Devin's cloud. Outposts is available on Pro, Max and Teams, which is a more specific arrangement than saying the whole product runs locally.

Claude's pricing page lists Pro at $20 billed monthly or $200 billed annually, Max at $100 per month and Ultra at $200 per month, excluding applicable tax. Subscription usage is limited, and API access uses token billing. Hosted Code Review is billed separately through organization usage credits.

Devin's pricing page lists Free at $0, Pro at $20/month and Max at $200/month, with usage allowances and purchasable on-demand credits. Its self-serve guide and pricing page disagree about whether Teams' $80/month is a base charge or a minimum. Confirm the team bill and automation funding before committing recurring work to that plan.

Claude Code and Devin, point by point

Both products answer the same questions in the table below. The source and check date beside each explanation let you inspect preview status and product-specific limits.

Both columns answer the same questions from each company's public pages. Yes means the company publishes the capability the question asks about, Partly means something narrower, No means it says it does not offer it, and Not published means its public pages do not describe it. Every answer links to its source and shows when we last checked it. The oldest check on this page was November 13, 2026.

Claude Code and Devin compared, with the source and check date of each answer
What you compare Claude Code Devin
Goals Claude Code Yes

The /goal command gives Claude a measurable completion condition, such as passing tests or an empty issue backlog. A separate model checks the condition after each turn and tells Claude to continue or stops the session's goal.

code.claude.com/…/goal.md
Devin Partly

Devin can coordinate work toward a requested outcome, such as improving modules below a test-coverage threshold. Its docs do not describe an owned goal with a tracked metric and ongoing reviews against its target.

Leads Claude Code Yes

Projects give Claude a coordinator role that starts and tracks threads, while experimental agent teams have a lead that assigns tasks and synthesizes results. Projects are public beta for Pro, Max and Ultra, with a gradual rollout.

Devin Yes

A coordinator session scopes a large job, delegates to managed Devin sessions, monitors progress, resolves conflicts and compiles results. This coordinates the delegated project rather than establishing a permanent team lead.

Tasks Claude Code Partly

Experimental agent teams share tasks with agent owners, dependencies and completion states, and Projects retain work in threads. Its docs do not describe the same work item being owned interchangeably by a person or an agent.

Devin Yes

Through Linear, tickets can be assigned to Devin and started with a default skill. Devin posts an activity feed, syncs its plan and accepts follow-up messages on the ticket's agent session. The work record lives in Linear.

docs.devin.ai/integrations/linear.md
Recurring work Claude Code Yes

Research-preview scheduled tasks run from recurring schedules, HTTP API calls or GitHub events. Each run opens a separate cloud session, with its history available from the saved task.

code.claude.com/…/routines.md
Devin Yes

Automations start Devin sessions on hourly, daily, weekly or custom schedules and from outside events, including GitHub checks, Linear updates and authenticated webhooks.

Workflows Claude Code Yes

Dynamic workflows put loops, branches and agent stages in a JavaScript script that you can read and rerun. The script controls what runs next rather than leaving each orchestration decision to the conversation.

code.claude.com/…/workflows.md
Devin Yes

Dynamic Workflows use deterministic Python scripts to order agents and pass structured results between stages. Calls are recorded, and interrupted runs replay completed results while unfinished work runs again.

Autonomy Claude Code Yes

Local permission rules allow, ask or deny tool actions, with approval prompts in the session. Public-beta Projects list threads needing approval under Waiting on you, although cloud threads do not apply repository permission rules.

Devin Yes

Devin CLI lets you allow, deny or require approval for commands, file access and Model Context Protocol tools. Approval prompts let you grant access once, for a session, for a project or globally.

docs.devin.ai/…/permissions.md
Learned autonomy Claude Code Partly

The /auto-mode-setup command drafts environment and rule entries from project configuration and commands in recent sessions, for you to accept. Its docs do not describe learning permission-change proposals from your approval history.

Devin Not published

Its docs describe memory from corrections and configurable approval rules, but do not describe Devin proposing permission changes from your approval history.

Memory Claude Code Yes

Auto memory keeps corrections and preferences across local sessions in files you can read, edit or delete. Public-beta Projects also have editable project memory that later cloud threads read.

Devin Yes

Personal memory carries preferences and corrections into later sessions. You can browse notes and ask Devin to change or remove them. Memory is personal within an organization, and automation sessions do not read or write it.

Planning Claude Code Yes

Plan mode researches a request and writes a proposed plan before implementation. You can revise the plan in your editor or ask Claude to keep planning before approving execution.

Devin Yes

Ask Devin lets you discuss your codebase and scope implementation before coding. It turns that conversation into a context-rich prompt for an Agent session, whose progress remains visible in the conversation.

Intake Claude Code Yes

An HTTP POST can send an incoming alert to a saved Claude Code task whose prompt tells the agent how to investigate it. GitHub events can also start these research-preview cloud runs.

code.claude.com/…/routines.md
Devin Yes

Automations route Slack messages, tickets and webhook events into new Devin sessions, existing sessions or a triage monitor. Conditions and optional preflight checks decide which events start work.

Connectors Claude Code Yes

MCP servers connect external tools and APIs. Research-preview scheduled tasks accept HTTP API triggers and GitHub events, and let you select which connectors each run can use.

Devin Yes

Devin CLI connects custom Model Context Protocol servers to external tools. Automations offer native service triggers and incoming HTTP webhooks with configurable conditions.

Models Claude Code Yes

The opusplan setting automatically uses Opus for planning and Sonnet for execution. Code Review also chooses its own Claude models rather than asking administrators to select one.

Devin Yes

Adaptive in Devin CLI evaluates each request and selects an underlying model, routing simpler tasks to faster models and complex tasks to more capable ones. You can select a specific model instead.

docs.devin.ai/cli/adaptive.md
Cloud or your own computer Claude Code Yes

Claude Code runs locally in your terminal and editors or in cloud sessions. Projects can send a thread to your computer through Remote Control when it needs local tools or files.

Devin Yes

Devin offers cloud sessions and Outposts for command execution, file edits and repository access on your own machines. Outposts keeps inference and planning in Devin's cloud and is available on Pro, Max and Teams.

docs.devin.ai/…/overview.md
Your own subscription Claude Code Yes

You can sign in with a Claude subscription, supply an Anthropic API key or configure supported cloud-provider credentials. The published connections cover Claude usage rather than ChatGPT or GitHub Copilot subscription allowances.

code.claude.com/…/authentication.md
Devin Partly

Pro, Max and Teams can connect personal ChatGPT Go, Plus or Pro subscriptions for eligible GPT requests. Other models and cloud machine time still use Devin funding, and Free and Enterprise accounts are excluded.

docs.devin.ai/…/chatgpt.md
Pricing Claude Code Yes

Claude Code supports token-based API billing and subscriptions with usage limits. Pro costs $20 billed monthly or $200 billed annually, Max $100 per month and Ultra $200 per month, excluding applicable tax.

Devin Yes

Free is $0. Pro is $20/month and Max $200/month, with usage allowances and purchasable on-demand credits. The pricing page and self-serve guide disagree on whether Teams' $80/month is a base charge or a minimum.

Correctness Claude Code Yes

The /goal evaluator checks a stated completion condition after each turn, such as tests passing, and sends unmet work back. It judges the conversation evidence and does not independently run commands or read files.

code.claude.com/…/goal.md
Devin Yes

Devin builds a focused test plan from a code change, executes it through the browser and sends a recording. Beta custom reviewers also check pull requests against written instructions, with configurable passing conditions.

For software work

Claude Code and Devin compared, with the source and check date of each answer
What you compare Claude Code Devin
Code review rules Claude Code Yes

Code Review flags changes that violate CLAUDE.md instructions and accepts review-specific rules in REVIEW.md. The hosted service is available on Team and Enterprise plans.

Devin Yes

Devin Review reads repository instructions such as REVIEW.md and AGENTS.md. Beta custom reviewers check pull requests against your instructions or skills and produce attributed findings.

Adversarial review Claude Code Yes

Research-preview ultrareview launches a separate fleet of reviewers in a cloud sandbox, with findings independently reproduced and verified. Dynamic workflows can also have independent agents adversarially review each other's findings.

Devin Partly

Beta custom reviewers can add focused checks alongside Devin Review, for example API compatibility. Its docs do not establish that these are independent second reviews specifically triggered by risky changes.

Risk rating Claude Code Not published

Its docs describe no overall risk rating for a revision before merge. Code Review ranks individual findings by severity and leaves the merge decision with your team.

Devin Partly

Beta Autoapprover evaluates intent, code quality and incident risk before approving a commit or leaving it for human review. Its published result is an approval decision with reasoning, not an overall merge-risk rating.

Pull requests Claude Code Yes

Public-beta Projects open pull requests from cloud threads and watch them with auto-fix. Threads push fixes for failed checks and address review comments before reporting the pull request ready for you.

Devin Yes

Devin creates GitHub pull requests, responds to review comments and pushes branches. Repository permissions control which operations it can perform.

docs.devin.ai/integrations/gh.md

A fact we cannot find or cannot check is left out, and prices can change after the date shown. Task Machine publishes this comparison and is an interested party. Claude Code and Devin are named only to identify them, and there is no relationship between the companies. Has something changed? Tell us.

Who each one suits

Pick Claude Code if local coding, editable memory, scripted workflows and its independent review tools fit your repository process. Its scheduled tasks are research preview and individually owned, while Projects availability and cloud permissions differ from local sessions.

Pick Devin if a Linear ticket handoff, repository delivery and browser-test recordings fit how your team accepts changes. Its recurring work needs deliberate guidance because automation sessions do not use personal memory, and the public Teams billing descriptions conflict.

Where Task Machine fits

Task Machine keeps recurring work under a stable root Task, with each occurrence's status and history separate, as its schedule documentation describes. People and agents can own Tasks, and corrections become inspectable memory scoped to the agent and its work. Task acceptance requirements and Workflow verifier feedback give the team a review model beyond an execution status.

Task Machine is the wrong pick if the requirement is simply to keep Claude Code's coding interface or Devin's existing ticket-to-delivery process. Inspect Task Machine vs Claude Code and Task Machine vs Devin, then trace one recurring repository job from its saved instruction to the evidence a human will accept.