Field notes from inside the work — GEO, agent deployment, evals, and what actually survives contact with production. Written by the engineers doing it, not a content team.
A software factory industrializes software production — a quality-gated line of senior engineers and AI agents, not bespoke projects. Here's what that means, how it differs from an AI factory and a dev agency, and why the mid-market should hire one.
Getting cited by AI answer engines
8 articlesWhat AI agents are, and which one you need
8 articlesHow to buy AI without getting burned
6 articlesWhat AI actually costs, and what it returns
4 articlesKnowing whether you are ready to build
3 articlesKeeping a live AI system working
Four cost centres, and the model is the smallest one. What an AI product costs to get in front of real users, what moves the number, and the line most budgets forget entirely.
ReadAI systems don't crash. They decay — quietly, while every dashboard stays green. The five ways a live system degrades, why nobody notices for weeks, and what the job of operating one actually is.
ReadThe formula is simple arithmetic. The hard part is being honest about four inputs — and three of them are routinely inflated by the person building the business case.
ReadYou asked the question your buyers ask, and a competitor got named instead. Five specific asymmetries explain almost every case — and four of them are fixable in weeks.
ReadA runbook you can execute this afternoon with a spreadsheet and four browser tabs. What to ask, what to record, and the three mistakes that make a DIY audit useless.
ReadNot what one agent costs — what the whole line item should be. How to size an annual AI budget across audit, build, and operate, and the split that survives a second year.
ReadThird vendor, same starting line. Why each switch resets you to zero, what the previous engagement was supposed to leave behind, and how to make this one the last.
ReadA scored diagnostic across five operational dimensions that tells you whether you can ship and what to fix first. What a good one contains, what it can't tell you, and when to skip it entirely.
ReadThree phases, each ending in something that runs rather than something that reads. Why the order can't be rearranged, and what goes wrong when a phase gets skipped.
ReadSeven entries, each answering a question someone will ask at an inconvenient hour. The document that turns "we have an AI system in production" into something a person can actually be on call for.
ReadThe inventory to collect, the acceptance test that proves the handover worked, and the four things that always go missing. Do this properly once and you never repeat the vendor reset.
ReadNot the vendor's tricks — your own. Seven internal line items that never appear in any quote, why the biggest one is your team's time, and how to price them before you approve anything.
ReadFive tells that show up in the first meeting, long before any code. Each one is a question you can ask out loud — and the answers tell you whether to fund it, reshape it, or kill it.
ReadA technical checklist you can audit this week. What has to be true of your pages before a model can retrieve, quote, and verify you — and which of the five almost every site fails.
ReadYour Zapier flows and RPA bots aren't obsolete — they're cheaper, faster, and more predictable than an agent for the work they fit. The line is exception rate, and here's how to find yours.
ReadAutonomy isn't a single setting. Four supervision patterns, the two questions that decide which one a task gets, and why the answer is usually different for different steps of the same workflow.
ReadYou've decided to build custom. The remaining question is who staffs it — and the honest answer depends on two things: whether AI is core to your product, and who's operating it in year two.
ReadVendor shortlists are increasingly being drafted by an assistant before anyone contacts you. You can find out in five minutes whether you're on them — and most companies have never checked.
ReadReview grids and roundup posts earned their power by owning the click. As the click collapses, that power doesn't disappear — it moves somewhere less visible, and it changes who you have to influence.
ReadA one-page machine-readable summary of your site, in Markdown, at /llms.txt. What it is, what it isn't, and an honest answer about whether it does anything yet.
ReadReadiness isn't enthusiasm — it's five specific conditions. Here's the checklist we run before we agree to build, and what to do when you only meet three of them.
ReadThe headline number isn't the price — it's the entry fee for an open-ended meter. How the trap is built, the five contract clauses that give it away, and what to ask for instead.
ReadThe questions that separate a firm that ships from one that demos — what each is really testing, and what a bad answer sounds like. Ask ours the same ones.
ReadProduct-market fit means the market wants what you built. Answer-Market Fit means the market's AI assistants default to recommending you. Here's what the concept means, why it's the new bar, and how to measure and reach it.
ReadThere's no standard sticker price for GEO — but there is an honest answer. The three ways it's priced, what actually drives the number, and exactly what we charge.
ReadShould you build an AI agent in-house or buy a finished one? The honest framework — with the data on why most internal builds stall, and where each path actually wins.
ReadA chatbot answers; an agent acts. The real distinction — autonomy, tools, and multi-step work — and how to tell which one your workflow actually needs.
ReadYes, you can deploy AI agents on PHI workflows — if the architecture is built for it from day one. BAAs, minimum necessary, audit trails, and the deployment patterns that pass review.
ReadCitation rate is the share of your category's buying queries where an AI assistant names you. Here's how to define it, measure it honestly, and what 'good' looks like by market position.
ReadHow a support agent actually reaches majority autonomous resolution without torching CSAT: the tier design, the escalation logic, and the eval discipline — with real deployment numbers.
ReadWhen a buyer asks an AI assistant for a vendor, the answer isn't ranked — it's assembled. Here's what the engines actually retrieve, what they trust, and how to become the company they cite.
ReadStraight numbers with sources: what a production AI agent costs to build and run, why the token bill is the cheapest line in the budget, and where teams actually overspend. From a firm that publishes its prices.
ReadModel, orchestration, retrieval, evals, observability, guardrails — the six layers every production agent needs, what we use at each layer, and where teams cut corners they regret.
ReadSearch is splitting into two motions — the blue-link index and the answer engine. This is the field guide we use to make a company the source an LLM quotes, not the one it skips.
ReadPilot purgatory is an engineering problem, not an ambition problem. Here are the eval, ownership, and rollback gates that separate a demo from a deployed agent.
ReadA model you can’t measure is a model you can’t trust in production. How we build evals before we build the agent.
ReadWhat it actually means to embed engineers in your operation — and why it beats a deck-and-walk-away consultancy for the 20–500 person company.
ReadSupport triage, onboarding, and renewal risk — ranked by time-to-value from real deployments.
ReadWhat changed in production AI this month. The signal, pulled from inside real deployments — not the hype feed.
Unsubscribe in one click. We never share your email.
The agent patterns, model releases, and tooling that actually reached production this month.
Eval results and benchmarks from real deployments — what moved, what regressed.
GEO shifts: who answer engines started quoting, and why.
If you’d rather see it work than read about it, a 30-minute discovery call is the fastest path.
