Devin is a cloud coding agent.
AM lives on hardware you own.

01 — Cloud privacy risk

Your code is on their servers — by default.

On Devin's Core and Team plans, every file you share, every codebase you connect, every command you run goes through Cognition's cloud infrastructure. Zero data retention and private VPC deployment are locked behind an Enterprise contract. If you are an individual developer or small team, you get none of those protections — the entry plans don't offer them. AM runs entirely on your own machine. Nothing leaves your hardware. Not your code, not your conversation, not your memory.

02 — Session-scoped memory

Devin forgets you after every session.

Devin has no persistent memory of who you are. Each new task starts from zero — no preferences, no project history, no understanding of how you think or what you have already tried. You re-explain context every time. AM builds a deepening model of your work over weeks and months. It remembers your projects, your decisions, your constraints, and your preferences. The longer you work with AM, the more useful it becomes — because it actually knows you.

03 — ACU cost model / failure rate

You pay whether it works or not.

Devin's Agent Compute Unit (ACU) model means every minute of agent runtime consumes budget — including the time Devin spends going down dead ends, misunderstanding requirements, or failing tasks. Independent testing found roughly a 15% success rate on complex tasks. The answer.AI review (January 2025) found 14 failures in 20 tasks. The Team plan starts at $500/month before ACU add-ons. AM is free and open source — you only pay Anthropic for API usage on tasks that actually run.

Is AM a good Devin alternative?

AM is built for people who want a personal AI worker — not a cloud coding agent. AM runs locally, builds persistent memory of your work and life, and is not limited to code. It handles research, writing, scheduling, and complex personal projects. If Devin's cloud-only model, session-scoped memory, or monthly subscription is a problem for you, AM was built to solve those problems.

Can AM write and run code like Devin?

Yes. AM can write, review, and iterate on code. Unlike Devin, it does so with full context of your project history, your preferences, and your architecture decisions — accumulated over time. AM is also not limited to code: it works across every domain you bring to it.

What are the biggest criticisms of Devin?

Independent reviews consistently found that Devin fails on anything beyond bounded, well-defined tasks. Answer.AI's month-long test (January 2025) found 3 successes and 14 failures across 20 tasks. The Register called it 'bad at its job.' The original Upwork demo was publicly debunked as a cherry-picked example that a human completed in 30 minutes. The SWE-Bench benchmark score of 13.86% was compared to zero-shot LLMs rather than other agents.

How does AM handle developer tasks specifically?

AM operates with full knowledge of your local environment — your repos, your tools, your terminal, your files. It can execute commands, read and write code, search documentation, and run tests. Because it knows your history, it doesn't start from zero every time. It knows which solutions you've already tried and which constraints you're working inside.

How much does AM cost?

AM is free and open source (MIT license). Install it on Mac, Windows, or Linux in one command. The only cost is your Anthropic API usage, paid directly to Anthropic. No AM subscription, ever.