Notes
Short field notes from certified engineers. We write about eval harnesses, MCP servers, Claude Code, and the hiring mistakes that keep pilots stuck in demo mode. Every piece is meant to help you ship, or to tell you honestly when you should not.
This is a studio blog, not a newsroom. We publish when we have a pattern that keeps showing up in repos: missing evals, tool schemas Claude will not call, agents with no budget ceiling, teams hiring generalists to learn Claude on the client's clock.
The notes sit in four clusters that match the work we sell. Production agents cover evaluation, guardrails, and cost. MCP and tools cover the contract between Claude and your systems of record. Claude Code covers repo conventions and enablement. Hiring covers when an audit is enough and when you need a six-week build.
If you want the work done inside your repository rather than described here, book a 15-minute discussion.
Eval harnesses, guardrails, cost ceilings, and agents that survive real traffic.
Production agents
A tool-using Claude agent without guardrails will be helpful in the wrong direction. Here is the refusal, injection, and review path we install before live traffic.
August 3, 2026/2 min read
Production agents
Prompt caching cuts Claude spend when the stable prefix is actually stable. Here is how we set cache breakpoints, ceilings, and evals so cost is a number, not a surprise.
July 13, 2026/2 min read
Production agents
LLM-as-judge is fine when the rubric is boring, structured, and checked by a human. Here is how we score Claude agents without turning CI into a $12 hobby.
June 22, 2026/2 min read
Production agents
Most Claude pilots stall because nobody can tell if quality got better. Here is the eval harness we install in week one, wired into CI, before the agent sees real traffic.
May 25, 2026/6 min read
Production agents
A golden dataset for Claude is a named, versioned set of real cases with invariants, not a folder of happy-path chats. Here is how we build one that CI can fail.
April 27, 2026/2 min read
Tool schemas Claude actually calls, and MCP servers that clear a security review.
MCP and tools
"We connected Claude to Salesforce" usually means one mega-tool and a security exception. Here is how we ship named MCP verbs that survive a CRM of record.
August 10, 2026/2 min read
MCP and tools
If Claude can refund, post, or export, it needs a review queue, not a braver system prompt. Here is how we design human checkpoints that agents actually wait on.
July 20, 2026/3 min read
MCP and tools
Tool use on the Messages API and MCP servers are not the same design. Here is how we choose which contract Claude should use for a given job.
June 29, 2026/3 min read
MCP and tools
An MCP server is a permissioned adapter, not a chatbot plugin. Here is the security review we run before Claude gets hands on Salesforce, SAP, or your internal APIs.
June 1, 2026/2 min read
MCP and tools
Claude skips, mis-fills, or over-calls tools when the schema is written for OpenAPI completeness instead of for a model. Here is how we design MCP tools that get used correctly.
May 4, 2026/5 min read
Repo conventions, CLAUDE.md, and enablement so your team keeps shipping after we leave.
Claude Code
A staged Claude Code enterprise rollout checklist covering managed policy, tool permissions, MCP, gateways, auditability, evals, cost controls, and adoption.
August 25, 2026/7 min read
Claude Code
If Claude Code can edit prompts and tools, CI must be able to fail those edits. Here is how we gate agentic coding the same way we gate application code.
August 17, 2026/2 min read
Claude Code
Enablement trains your team on your repo. Hiring adds a headcount. Here is how we choose, and why the exams are the syllabus rather than trivia.
July 27, 2026/2 min read
Claude Code
CLAUDE.md is repo policy for Claude Code, not a novel. Here is what we put in it so agentic coding stays inside your conventions after we leave.
June 15, 2026/2 min read
Claude Code
Agentic coding fails when the repo has no rules the agent can follow. Here is the convention set we install so Claude Code PRs look like your team's, not like a weekend plugin.
May 18, 2026/3 min read
Claude Code
A certified engineer opens a pull request inside seven days because the repo is ready, not because the model is magic. Here is the week-one path we use in live codebases.
April 13, 2026/2 min read
When to audit, when to build, and when certification is the hiring filter.
Hiring certified engineers
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
Hiring certified engineers
If the prompts, evals, and MCP servers do not live in your git from day one, you are renting a demo. Here is the IP rule we sign before week one, and what to reject in a vendor contract.
July 6, 2026/3 min read
Hiring certified engineers
Developer-certified engineers build the tools and agents. Architect-certified engineers decide what must not be built. Here is how we staff both so you do not get a clever system nobody can govern.
June 8, 2026/3 min read
Hiring certified engineers
A two-week audit, a six-week production build, and a quarterly embed solve different problems. Here is how we choose so you do not buy a rebuild when you needed a diagnosis.
May 11, 2026/4 min read
Hiring certified engineers
Most failed Claude projects fail for the same four reasons. Anthropic's exams exist because those reasons are learnable. Here is how we hire and staff against them.
April 20, 2026/3 min read
Subscribe in a reader via RSS.
