Blog

AI in production, in practice.

Real workflows, measured outcomes, and the lessons from an 18-month internal rollout - no hype.

All Articles

Latest from our team

Overhead flat-lay of five printed requirement pages on a slate desk labelled Intent, Evaluation, Failure Modes, Safety and Cost, linked by an ink line
Agentic Engineering

What Your PRD Is Missing When the Product Is Probabilistic

"The assistant returns an accurate summary of the ticket." That sentence passes review, gets signed off, and cannot be verified, because there is no single run…

Read article →
A monitor showing an AI assistant answer with visible source citations, an open reference panel, and approve, edit and reject controls
AI Adoption

The AI UX patterns that decide whether anyone uses your feature

AI UX patterns that earn trust: transparency, explainability, confidence indicators and review loops, plus a build order ranked by real stakes.

Read article →
A glass office wall with a hand-drawn LLM architecture diagram in marker, labelled orchestration, model, retrieval and observability, one box crossed out and redrawn
Agentic Engineering

The AI Product Architecture Decisions You Cannot Cheaply Undo

Architecture diagrams for AI products have converged on the same picture. Four stacked boxes, orchestration over model over retrieval over observability, with…

Read article →
A cork planning board with five pinned phase cards reading discovery, MVP, production readiness, scaling and enterprise rollout with week ranges
AI Adoption

How Long It Really Takes to Build an AI Product

Every published answer to "how long does it take to build an AI product" was measured on something that isn't an AI product. The numbers are real: 20 to 40…

Read article →
Overhead flat-lay of five printed job specifications on slate, labelled Software, AI, Agent, Evaluation and Product, joined by an ink line
Agentic Engineering

What Skills Your AI Development Team Actually Needs

Five people answer the same job ad. A researcher who fine-tunes vision models. An engineer who has shipped three retrieval systems into production. A data…

Read article →
A monitor showing a review queue of identical rows with amber status pills, the list running past the bottom of the screen
AI Adoption

Your Engineers Have Copilot. Your Delivery Speed Did Not Change.

The same tool has been measured making developers 55% faster and 19% slower, and both studies were carefully run.

Read article →
A monitor on a wooden desk showing an operations dashboard where every status tile is green and every trend line is flat
Agentic Engineering

Your AI Feature Passed Every Eval and Is Failing Right Now

Every test passed and the feature has been wrong for six weeks. The six signals a production AI feature needs, which of them should page a human, and who owns the response.

Read article →
Glass wall in an engineering room with a hand-drawn five-part acceptance criterion in marker: metric, threshold, sample, reviewer, fail action
Agentic Engineering

How to Test AI Features When Outputs Are Non-Deterministic

Somebody in your organization is signing off on releases using a definition of "passing" that stopped describing the system. If you run engineering or QA at a…

Read article →
AI Adoption

How Much Does AI Product Development Cost in 2026

The model API bill is usually the smallest of seven budget lines. What an AI product costs to build and to run in 2026, with realistic ranges and the driver behind each one.

Read article →