Hiring certified engineers
Six weeks to a production Claude agent
Prototype by week two, live traffic by week six. Here is the path we run on a Production Agent Build, and what has to be true on day zero or the calendar is fiction.
August 24, 2026/3 min read/Claude Certified Engineers
A six-week production Claude agent is a scoped build, not a personality trait of the model. We run it as Production Agent Build: one job, real tools, retrieval if the job needs it, budget ceilings, guardrails, and a review queue. Prototype by week two. Live on real traffic by week six. Fixed fee, $30k to $65k.
The calendar only holds if week 0 is honest. If "one agent" is actually a platform, or SSO takes twelve days, you did not buy a six-week build. You bought a story.
The week-by-week path
This is the same timeline we publish on the site.
Week 0. Scope. One production outcome, one success metric, access provisioning, and a written eval rubric. If you cannot name the job in a sentence ("resolve invoice exceptions that already have a PO"), stop. Certification is supposed to make Architect-level people refuse fuzzy scope.
Week 1. Embed. Repo, CI, Slack, standups. A certified engineer opens their first pull request inside seven days. The eval harness starts here, before live traffic. Not after the demo.
Week 2. Prototype. A working Claude workflow on your stack against your real data, behind a feature flag. If week two is a slide, week six will be a slide.
Weeks 3 to 5. Harden. Tool integration, guardrails, caching, eval gates in CI, cost ceilings, audit logging. This is where MCP verbs get split, human queues land on refunds and posts, and token budgets become a number you can forecast.
Week 6. Production. Live on real traffic with tracing, review queue, and quality metrics reporting daily.
Handoff. Runbook, eval suite, and pipelines in your repo. Your team trained. IP fully yours, because it never lived anywhere else. See IP ownership.
What we will not pretend is six weeks
- Five agents, or "ChatGPT but for the company."
- Replacing Salesforce.
- A model bake-off with no production path.
- Greenfield infrastructure when your team cannot merge to main.
Those engagements exist in the world. They are not this offer. If you need a diagnosis of an existing system first, buy the two-week audit. If you need someone after week six, buy Embedded or enablement.
Day zero failures that blow the clock
Access. Flaky CI. No owner on your side. Data you cannot use in staging. An InfoSec review that starts in week five because nobody put the MCP server on a diagram.
We hold a small number of slots each quarter so two weeks from signed scope to first PR stays typical. We cannot hold the six-week end date if week 0 never finishes.
The outcome we are hired for looks like the quotes on the homepage: a demo that sat for nine months, then real customers by the end of the quarter. That only happens when the first PR is real, the eval is a gate, and Architect-level review is allowed to cut tools. The model is not the scarce part. The path is.
Questions
- How long does it take to put a Claude agent in production?
- Six weeks on a Production Agent Build: prototype by week two, live traffic by week six, if scope is one outcome and access lands on day zero. Two weeks from signed scope to first pull request is typical.
- What happens in week one of a Claude build?
- The engineer embeds in your repo, CI, and Slack, opens a first PR inside seven days, and lands the start of the eval harness before the agent sees live traffic.
- What if we already have a nine-month demo?
- If you cannot score it, start with a two-week audit. If you will not harden it, skip to a six-week build and do not spend a month polishing the demo.