Book a Discovery Call

Book a Discovery Call

Book a Discovery Call Contact us

Insights

Field notes from production AI.

No hot takes, no hype cycles. Practical writing on the engineering that actually gets AI into production and keeps it there, from the team doing the work.

Featured · Delivery

Why most AI projects die between the demo and production

The demo is the easy 20%. The reason AI initiatives stall is almost never the model; it’s the integration, evaluation, observability, and cost work that turns a prompt into a system.

April 22, 20266 min read

Latest

More from the team.

Platform & reliability

AI observability in production: what to measure before it breaks

Logs tell you that a request failed. AI observability should tell you why quality, latency, or cost changed, across models, retrieval, tools, and the full user journey.

July 13, 20267 min read

An AI production-readiness checklist

Before you put an AI feature in front of users, run it through the checks that separate a demo from a system: evaluation, integration, observability, cost, and safe failure.

June 3, 20267 min read

How to evaluate an LLM feature before you ship it

You can’t improve what you can’t measure. A practical approach to evaluating LLM features using golden sets, offline scoring, and online guardrails before and after launch.

May 21, 20266 min read

Agentic workflows in production: what actually works

Most ‘agents’ that reach production are really well-structured workflows. When to reach for an agent, when to use a pipeline, and the patterns that survive real traffic.

May 7, 20267 min read

RAG is a retrieval problem, not a prompting problem

When a RAG system gives wrong answers, teams reach for the prompt. The fix is almost always upstream: what you retrieved, how you chunked it, and whether you can even see why.

March 11, 20265 min read

Evaluations are the only thing between you and silent regressions

A prompt tweak or model upgrade can quietly degrade quality, and you won’t know until a customer tells you. Evaluation harnesses turn that invisible risk into a number you can act on.

February 4, 20265 min read

Cost & reliability

Model routing: cutting AI cost without cutting quality

Sending every request to your most expensive model is the default, and it’s a margin problem waiting to happen. Routing, caching, and instrumentation fix it without users noticing.

January 15, 20265 min read

Designing agents you can actually operate

Autonomy is easy to demo and hard to run. Operable agents come from explicit control flow, typed tools, and traceability. Bounded autonomy beats open-ended cleverness in production.

December 9, 20256 min read

Shipping AI under the EU AI Act: governance without the theater

The AI Act is now a delivery concern, not just legal overhead. The good news: most of what it asks for (documentation, risk management, and traceability) is also just good engineering.

November 6, 20256 min read

Let’s talk

Prefer to talk it through?

If something here maps to a problem you’re facing, a discovery call is the fastest way to a concrete answer.

Book a Discovery Call See the work