Devin is now generally available

Original Article ↗ Hacker News Discussion ↗

An AI coding agent called Devin has entered general availability with a $500/month team subscription, promising to act more like an autonomous teammate than a traditional code assistant. Commenters weigh its claimed strengths—full-project workflows, API integrations, PRs, and Slack-based collaboration—against concerns about speed, reliability, benchmarks versus Claude/o1/Copilot, pricing, and data rights. Many see potential in tool-using agents that operate via full VMs, but skepticism remains high due to limited public evaluation, opaque model details, and trust issues from earlier marketing claims.

What Devin Is and How It Works

  • Framed less as “Copilot-style autocomplete” and more as an autonomous assistant you assign projects to.
  • Runs in a cloud VM (Linux, with shell, browser, file system, and internet) and can open docs, run code, debug, and create PRs.
  • Asynchronous workflow: users often kick off multiple tasks, let it work, then review output and PRs later.

Comparison to Claude / ChatGPT / Cursor / Copilot

  • Some say comparison to raw models (Claude 3.5, o1) is misleading because Devin’s value is in tools + workflow, not just code generation.
  • Others insist comparison is valid: if it can’t solve problems that stump Claude/o1, the high price is hard to justify.
  • Several want head‑to‑head, up‑to‑date benchmarks vs Claude 3.5/3.6 and o1 Pro; current public comparisons are viewed as outdated or incomplete.

Capabilities and Use Cases

  • Reported strengths:
    • Large, complex refactors and migrations touching many files, beyond simple search/replace.
    • Multi-document writing projects (proposals, legal/feasibility docs, training materials).
    • API integrations and long-tail “annoying” debugging, especially with messy or poorly documented APIs.
  • Can in principle work in any language since it reads compiler errors; mixed feedback on C/C++ quality.
  • Can handle very large repos, but initial scanning and knowledge building can take hours to days.

Performance, Speed, and Resource Use

  • Users note it used to be quite slow; recent updates reportedly 3–4x faster, but some still want a “speed/demo” mode.
  • Uses credits (ACUs); rough guidance: small frontend task ≈ 1–2 ACUs; VM burns ACUs while actively working.

Pricing and Access

  • Current visible offering: $500/month team tier with included credits; lower-cost $50 personal tier exists but is closed to new signups.
  • Some want pay-as-you-go or cheaper, constrained tiers to experiment; $500 just to try is seen as a barrier.
  • Complaints that pricing isn’t obvious early in the signup flow.

Data, Trust, and Legal Terms

  • Questions about data handling, training on user code, and third-party model providers; answers in thread are partial/unclear.
  • ToS clause granting broad license to customer data is debated; some see it as standard SaaS language, others as worrisome.
  • A subset of commenters distrust the company due to earlier accusations of misleading marketing and the lack of independent benchmarks or public testing.

Developer Experience and Future of Work

  • Some see Devin as a “superpower” that offloads repetitive maintenance and boilerplate.
  • Others worry delegating complex integration and architecture work to a black box erodes deep understanding and craftsmanship.
  • Questions arise about long-term impact on junior/mid-level developer roles; outlook is mixed and largely unresolved.

Naming and Branding Reactions

  • Side discussion about using common first names (“Devin”, “Alexa”) as product names; some find it inconsiderate or cheesy, others don’t care.