Pular para o conteúdo

Cheesecake Labs is now a Select partner in Anthropic’s Claude Partner Network

Select Partner — Blog | | Cheesecake Labs

4 min read in AI Implementation

Published
Updated
Summary
  • Cheesecake Labs has been promoted to the Select tier of the Services Track in Anthropic's Claude Partner Network, a tier that requires certified engineers, joint customers running Claude in production, and a public customer story.
  • Select partners are verified by Anthropic and listed in the public Claude Partner Hub directory, where companies can see each firm's tier, certified team, deployments and references.
  • The post notes that many companies complete a Claude pilot, but stall when connecting Claude to real systems, testing it, setting guardrails and getting people to use it; the firm says closing that gap is its focus.
  • Its process is production-first and governance-built-in, and it uses Claude Code and multi-agent review on pull requests, but a person makes every approval decision.

Cheesecake Labs has been promoted to the Select tier of the Services Track in Anthropic’s Claude Partner Network. Select is the tier where, in Anthropic’s words, partnership begins, and it isn’t granted for signing up. It’s earned on evidence: certified engineers, customers running Claude in production, and clients willing to say so publicly.

What Select means

Anthropic launched the Claude Partner Network in March 2026 and added the Services Track in June to help enterprises find the firms that can put Claude into production. Every firm is measured against the same published requirements, whatever its size. To reach Select, a firm needs:

  • At least 10 active Claude-certified practitioners

  • At least 2 joint customers with Claude deployed in production in the last 12 months

  • At least 1 public customer story

Anthropic verifies each partner’s standing, and Select partners are listed in the public Claude Partner Hub directory, where companies evaluating partners can see each firm’s tier, certified team, deployments and references.

Why it matters if you’re building with Claude

Most companies we talk to have already run a Claude pilot. The pilot usually works. What stalls is everything after it: connecting Claude to real systems and data, testing it against the cases that matter, setting the guardrails and review steps the business needs, and getting people to actually use it.

That gap between a promising pilot and a system a business can run on is the work we do. For Knapsack, it meant building Claude-powered agents that could be trusted inside the strict requirements of a financial environment.

Select status gives you a simple, independent check: Anthropic has verified that our team is certified and that our Claude work is live with customers, not just in demos.

How we build with Claude

  • Production first. We scope every engagement around a workflow that has to run in production, with integrations, evaluations and monitoring planned from the start rather than added at the end.

  • Governance built in. Access controls, human review points and audit trails are part of the design, so security and compliance teams can sign off.

  • Built by people who use it. Our own teams work with Claude every day, so our recommendations come from experience, not from slides.

Inside our process: Claude writes and reviews, people decide

On one of our current client projects, Claude Code writes the code in each pull request and the review runs on Claude too. The rule is simple: a person makes every approval decision. Here’s how a feature moves from ticket to release:

  1. Spec before code. Each ticket becomes a technical design and refinement document. Claude implements from that, not from a loose prompt.

  2. Acceptance-criteria check. Before the PR opens, the diff is checked against the ticket’s acceptance criteria. Missing coverage and behavior nobody asked for become questions for the author.

  3. PR gate. Pull requests stay in draft until a technical demo to the team, and each one states what was tested and what wasn’t.

  4. Multi-agent code review. Six review agents run in parallel (security, quality, architecture, performance, requirements and regression), and an independent verifier confirms each finding before it becomes a comment on the PR.

  5. Side-by-side QA. The branch and the base run together. Each acceptance criterion is proven in the browser and against the API, screens are checked against the designs, and changed lines go through mutation testing.

  6. Human sign-off. The reviewer records a verdict for every criterion (passed, failed, diverged, or not applicable with a reason) plus anything left uncovered. The approval is theirs.

  7. Staging acceptance. After merge, the author validates the feature in staging against the same criteria before it ships.

The whole process is encoded in versioned Claude Code skills and repository hooks, so it runs the same way on every pull request and no step can be skipped.

What’s next

We’re growing our certified team and adding more customer stories as current projects go live. Anthropic has also announced industry and use-case specializations for partners, which we plan to pursue in the areas where we already deliver.

If you have a Claude pilot that never made it to production, or a process you want to hand to agents, let’s talk.

See also

See more