Nori vs Devin: which AI engineer should a small team pick?

Quick answer

Both ship an AI engineer: Devin from Cognition, Jiro from Nori. Pick Devin if you want one managed agent with a $20 entry plan and you accept that the agent brain stays in Cognition's cloud. Pick Nori if you want Jiro or any harness, such as Claude Code, Codex or Gemini CLI, working for the whole team from Slack, schedules and webhooks, with shared memories and skillsets, on your own Claude or ChatGPT subscription, with flat plans and no token charges.

What is Devin?

Devin is Cognition's AI engineer: a hosted coding agent that takes a task, works on it in the cloud and returns the result for review. It suits teams that want a single managed agent from one vendor rather than a platform that runs several.

As of November 2026, Cognition's documentation establishes four things that matter for this comparison:

• Plans. Pro is $20 per month and Max is $200 per month. Teams has an $80 minimum. See self-serve billing.
• Deployment boundary. Cognition retains the agent brain even when you use Outposts, its option for running session machines elsewhere. See the deployment overview.
• Subscription reuse. Eligible personal ChatGPT plans can fund GPT usage, with exclusions. See paying with a ChatGPT plan.
• Memory. Automatic memory is personal, and it is excluded from automation sessions. See the memory guide.

Cognition is also growing by acquisition: the personal assistant Poke announces on its homepage that it is joining Cognition. This page makes no claim about Devin features beyond the documents listed under Sources.

What is Nori?

Nori is a cloud platform for coding agents, built for engineering teams. The whole team drives it from Slack, Discord, the web or the CLI. Cron schedules, one-time reminders and webhooks start the same kind of session a person would start, with run history and approval paths. Every session begins with the team's shared memories and org skillset, so the agent already knows how you ship.

Each session is an isolated cloud sandbox running the harness you choose: Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Droid, Amp or any ACP agent. You connect your own Claude or ChatGPT subscription, API keys or any LLM gateway. Nori never charges for tokens, so heavier use does not change what you pay Nori.

Meet Jiro

Jiro is the AI engineering teammate that Nori ships, and the closest Nori counterpart to Devin. It runs on Claude Code or Codex under the hood, on your own subscription or API key. It follows an opinionated engineering workflow: plan, write the failing test, implement, review. On its first day it reads your repositories, asks the team how you ship and sets up the workspace. See jiro.bot.

A typical request looks like this:

#payments-eng
@Jiro the refund webhook returns 500 when `currency` is null.
Reproduce it, add a regression test in tests/webhooks/test_refund.py, open a PR.

Jiro: Plan posted in thread. Failing test written, fix pushed.
PR: fix(refunds): handle null currency in webhook payload
Tests: 412 passed. Ready for review.

Nori vs Devin, head to head

The memory row matters more than it looks. Suppose a nightly job triages flaky tests. On Nori the scheduled session starts with the same shared memories a person's session gets, so a note such as "test_ledger_sync is flaky on Postgres 16, retry once before filing" applies at 02:00 as well as at 14:00. Devin's documentation says automatic memory is personal and does not apply to automation sessions, so plan to carry that context another way.

Enterprise deployment: where does the control plane run?

The control plane is the part that orchestrates sessions, holds transcripts and organization state, and runs the agent loop. It is a different question from where the session machines run. For background, see what an agent control plane is.

Devin's deployment documentation says Cognition retains the agent brain even with Outposts. Your machines can do the work, but the loop that decides what to do next stays with the vendor.

Nori Enterprise deploys the full stack, control plane included, into your own AWS VPC or Azure VNet, or on-prem, with your identity provider (SAML, SCIM, OIDC), LLM gateway and compute. On hosted plans, Nori gives each organization a dedicated, single-tenant control plane and data store, and every session runs in its own isolated sandbox.

When to choose Devin

• You want a $20 entry point. Devin Pro is $20 per month. Nori's paid plans start at $99 per month after a 30-day free trial.
• You want the agent and the usage bill from one vendor. Devin bundles the agent and its usage into Cognition's billing, so nobody on your team manages model credentials. Jiro needs a Claude or ChatGPT subscription, API key or gateway that you connect.
• You only use GPT models and hold an eligible ChatGPT plan. Cognition documents that such a plan can fund GPT usage, with exclusions.
• One person is the main user. Personal automatic memory fits a single engineer who mostly works interactively.
• A vendor-run agent brain is acceptable. If your security review has no objection to the agent loop running in Cognition's cloud, the deployment boundary is not a blocker.

When to choose Nori

• The whole team should share one agent setup. Requests from Slack, cron schedules and webhooks all start the same sessions, with the same shared memories and skillset.
• You want an AI engineer without token billing. Jiro plans, works test-first and reviews its changes on the subscription you already pay for.
• Your team already trusts a harness. Nori runs Claude Code, Codex, Gemini CLI and the rest as they are, so prompts, skills and habits built locally carry over.
• You pay for Claude. Nori uses a Claude subscription directly. Claude subscription support is not documented for Devin.
• You route models through a gateway. Nori works with your API keys or any LLM gateway.
• Scheduled work needs team memory. Shared memories apply to automation sessions on Nori.
• You need everything in your environment. Nori Enterprise runs the full stack in your cloud or on-prem.
• Your team uses Discord. Nori is driven from Discord as well as Slack.

How to switch from Devin to Nori

1. List what Devin does today. Write down each automation (trigger, prompt, repository) and the standing knowledge you rely on. Devin's automatic memory is personal, so ask each engineer for theirs.
2. Start the Nori trial. Thirty days, no card. Pick Jiro or the harness you want to standardize on.
3. Connect models. Sign in with your Claude or ChatGPT subscription, or add API keys or your LLM gateway.
4. Move instructions into the org skillset. Repeatable procedures become skills. Facts about the codebase become shared memories, for example: "Migrations live in db/migrate; never edit a merged migration."
5. Recreate automations. A nightly job becomes a cron automation such as the one below. Event triggers become webhook automations pointed at the event source.
6. Run both for a week. Connect Slack or Discord, give both tools the same five tickets, and compare the pull requests before you cut over.

schedule: 0 2 * * 1-5
agent: Jiro
repo: acme/payments-api
prompt: Triage tests that failed on main in the last 24 hours.
        Open one PR per root cause. Post a summary in #payments-eng.

For other options, see cloud coding agents compared, Nori vs Factory Droids and Nori vs Claude Tag. To see Nori on your own repositories, book a demo.

Frequently asked questions

Sources

Competitor facts were read from these vendor pages on November 11, 2026. Vendors change plans and features often; check the source before you decide.

• Devin docs: Self-serve billing (plans and prices)
• Devin docs: Enterprise deployment overview
• Devin docs: Paying with a ChatGPT plan
• Devin docs: Memory
• Poke: homepage (announcement about joining Cognition)
• Jiro, the agent Nori ships