Insights

Field notes from production AI.

No thought leadership. Just what we learn shipping agents, LLM apps, and the systems behind them — written by the engineers doing the work.

Scroll

16 essays · clear filters

All55 AI Agents8 Generative AI3 Engineering28 Enterprise16

Enterprise

What AI changes in procurement and security review

A vendor questionnaire assumes software that behaves the same way twice and a vulnerability that is either fixed or not. Neither assumption survives a probabilistic system — so four documents have to change.

Sep 12, 20268 min read

Enterprise

Who owns an AI system after it ships

Five jobs have to carry a name on the day an AI system goes live, and most launches cover all five with one person: whoever built it. Here is what the org chart actually has to look like.

Sep 12, 20268 min read

Enterprise

What breaks when AI meets a system of record

Connectivity is a week of work. Five other things break when AI meets the CRM, the ERP or the ledger — and four of the five are identity and reconciliation problems, not model problems.

Sep 10, 20266 min read

Enterprise

Enterprise AI adoption, measured

Vendor adoption numbers are enormous and government statisticians' numbers are small. Both are correct. Here are the measured figures, from sources you can open yourself.

Sep 10, 20267 min read

Enterprise

When you need an AI development partner, and when you don't

Five conditions make an outside AI partner the wrong hire, and all five are visible before you sign. Then the narrower set of cases where outside help genuinely pays — written by a firm that sells it.

Sep 10, 20267 min read

Enterprise

The five kinds of AI solution providers

Platform vendors, systems integrators, staffing firms, custom engineering firms and in-house teams. Each is right for a question the other four answer badly, and each has a characteristic way of failing. Written by one of the five, including where our own category is the wrong choice.

Sep 10, 20268 min read

Enterprise

Model drift: what changes underneath your AI system

Drift usually means accuracy quietly degrading. The bigger operational risk is blunter: the model you built on gets retired on a published date, and your requests start failing.

Sep 9, 20266 min read

Enterprise

Enterprise AI: build, buy, or assemble

It is almost never build versus buy. Nearly every production AI system is assembled — and the question that decides the architecture is which parts you cannot outsource.

Sep 9, 20266 min read

Enterprise

What an enterprise AI operating system actually includes

The phrase sounds like marketing. Here is the concrete list — four shared services a single AI feature almost never has, and the test for whether you have them.

Sep 9, 20266 min read

Enterprise

AI proof of concept to production: what a demo never needed

The four structural gaps a working demo never has to close, the industry's own numbers on how often that gap sinks the project, and the cheapest one to fix before launch.

Sep 3, 20266 min read

Enterprise

In-house AI team vs. AI development agency: how to decide

The hiring-time math that actually decides build vs. buy, the three questions that surface which one your project needs, and the one place a hybrid answer works.

Sep 3, 20266 min read

Enterprise

Onshore vs. offshore AI development: the honest tradeoffs

Why the classic offshore calculus doesn't transfer to AI work, what Deloitte's own outsourcing data shows about why companies are already reversing it, and the three things that actually need to sit onshore.

Sep 3, 20266 min read

← PrevPage 1 of 2Next →

Get each one as it ships

Prefer to get this through Search? Make us a preferred source on Google.

Let's put AI to work in your business.

A 30-minute call. You bring the workflow or the roadmap — we'll tell you what's feasible, what it costs, and what we'd build first.

Book a call→