An AI-native software factory. Audit → Build → Operate.
We don’t sell advice and walk away. We embed in your operation, build the production system, and stay accountable for the result. Three phases, one partner, and every phase ends in something that runs rather than something that reads.
Why one model, end to end
Most AI delivery fails at the seams. Strategy is done by one firm, the build is handed to another, and operations belong to whoever is left holding it, so context is lost twice and accountability is lost entirely. Audit → Build → Operate is one partner across all three.
Any consultancy can name three phases. The question is what each one produces. A phase that ends in a document has produced nothing that works, so each of ours ends in software: a pilot, a production system, a system that is still working a year later.
Three phases. What each costs, and what it ends in.
Every project follows the same model, so you learn it once and always know which phase you’re in.
Find the work worth automating.
Answer three questions with enough confidence to commit real money: which workflow, what will it cost, and will it pay. We map the workflow, model the ROI, and build a pilot of the thing we are proposing, because building it is the only way to find out what a build really involves.
Ends with: A pilot running on your own data, an ROI model, and a costed build plan.
Gate to the next phase: Go only if the pilot works on real data and the costed plan pays back. If it does not, you stop here with the answer, and the fee was the price of not building the wrong thing.
Already know the one workflow? The AI Quick Win Sprint ($8,000, 3 weeks) skips the diagnostic and ships it.
Ship the production system in your stack.
An embedded pod builds inside your systems, not alongside them: two-week sprints, a working demo every sprint, evals before deploy, observability from day one, guardrails and fallbacks. The output is either production agents at a flat fee each, or an embedded team when the work is broader than one system.
Ends with: A production system, its eval suite, the documentation and the code, all in your repository.
Gate to the next phase: Go live only when the system passes its evals on real cases, has tracing and alerting in place, and has a tested rollback path.
The $25,000 Sprint fee credits toward a Gigabit Agents build if you go ahead.
Stay on the hook.
Monitoring, eval regressions, prompt versioning, drift management, response to vendor changes, and a quarterly roadmap. AI systems decay rather than crash: models get deprecated, data drifts, someone changes the workflow. Operating the system is what keeps it worth what you paid.
Ends with: A system that keeps working, a monthly report against your numbers, and a quarterly review.
Gate to the next phase: Reviewed every quarter against the metric the system was built to move. If it stops earning its keep, we say so.
Light $3,000/mo (1 system) · Standard $8,000/mo (3 systems) · Heavy $20,000/mo (5+ systems or production agent platforms)
Each phase produces the input the next one needs.
| Phase | Answers | Skip it and |
|---|---|---|
| Audit | Which workflow, and will it pay? | You build the wrong thing well. |
| Build | Does it work unattended? | You have a demo, not a system. |
| Operate | Does it still work? | It decays, and you rebuild in about eighteen months. |
Build without Audit has no defensible target. Operate without Build has nothing to operate. And Audit without a committed Build is the most common shape in the market: the strategy engagement that ends in a deck, which is why so many companies have a careful AI roadmap and nothing in production.
We work where the work happens.
Forward-deployed is how the factory runs: engineers embedded inside your environment, close to the workflow, with their hands on your stack. The people who scope the work are the people who ship it and the people who stay accountable for it. Same sprint board, same Slack, same code-review bar. Your stack, your data, our engineers.
The reason it works is proximity. The context an AI build needs lives where the work happens: in the edge cases, the exceptions and the knowledge no specification captures. Put the engineer there and you skip the months of back-and-forth that stall most engagements. Where the role came from, and how it differs from consulting and staff augmentation, is in what is a forward deployed engineer.
It fits the 20-to-500-person company best: too small for a large consultancy that bills a steering committee, too serious for a demo shop that hands over a prototype and disappears. An embedded pod assembles in about two weeks and ships its first pull requests by day five. If you are weighing it against hiring or buying a product, the build vs buy vs embed comparison sets out the trade-offs.
Buying a capability, not renting one.
The code
In your repository from the first sprint, under your review rules.
The eval suite
The test cases that prove the system works, so any team can change it safely.
The documentation
How it is built, how to run it, and how to roll it back.
That list is the difference between buying a capability and renting one. Without it, each new vendor restarts from a blank page because the last one left nothing behind. With it, any competent team, yours included, can pick the system up.
Where Ascent fits.
One programme sits beside the sequence rather than inside it: Ascent, our GEO and AI-visibility programme. Audit → Build → Operate makes your operation work; Ascent makes buyers able to find you when they ask an AI assistant. They are independent, and you can run one without the other, but they share the same shape: a measured baseline, a build, and monthly accountability to a number rather than to activity.
What we put first
Engineering before content
We were engineers first.
Pricing before pitching
The number is on the page.
Evals before deploy
We prove it works before you trust it with production.
Production before pretty
A working system beats a beautiful mockup.
Adoption before scope
One workflow in production beats ten in a plan.
Honesty before optimism
We tell you what won’t work, too.
The model, answered
What is the Audit → Build → Operate model?
A three-phase AI delivery model run by one partner end to end. Audit finds the workflow worth automating and ends in a working pilot plus a costed build plan. Build ships the production system in your stack and ends with the system, the eval suite, the docs and the code in your ownership. Operate keeps it working through monitoring, eval regressions, drift management and vendor-change response.
How much does each phase cost?
Audit is the AI Transformation Sprint: $25,000 fixed for 2 weeks, or the AI Quick Win Sprint at $8,000 when you already know the workflow. Build is Gigabit Agents, from $8,000 · flat, per agent, or Embedded AI Teams at $3,000–$7,500 / engineer / mo. Operate is Managed AI Operations at $3,000–$20,000/mo. The Sprint fee credits toward a Gigabit Agents build.
Why should an AI audit phase be paid?
Because free discovery isn't free: it is priced into whatever comes next, and it creates switching cost rather than clarity. A paid, short diagnostic that ends in a deployed pilot and a costed plan gives you a real number and something that runs, and our Sprint fee is credited toward the build it scopes.
What happens if you skip the operate phase?
The system decays, and you rebuild within roughly eighteen months. AI systems degrade rather than crash: models get deprecated, data drifts, and workflows change underneath them, so an unoperated build quietly produces worse outcomes than anyone realises until something forces a look.
What does forward-deployed mean?
Forward-deployed means the engineers who scope the work sit inside your operation, close to the workflow, with their hands on your stack, and they are the same people who ship it and stay accountable for the outcome. It is the opposite of a strategy deck delivered from a distance or an offshore team briefed through a ticket queue.
Who is the forward-deployed model best for?
The 20-to-500-person company that is too small for a large consultancy billing a steering committee and too serious for a demo shop that hands over a prototype and disappears. An embedded pod assembles in about two weeks, ships first pull requests by day five, and owns delivery from audit through production.
How is this different from a consultancy's three phases?
Every phase ends in working software you own. The audit ends in a deployed pilot rather than a findings deck, the build ends with code and evals in your own repository rather than on a vendor platform, and operations are measured against your KPIs rather than tickets closed.
Do I have to buy all three phases?
No. Each phase ends in something you can keep and take elsewhere. You can start with a single AI Quick Win Sprint, or bring us in only to operate a system another team built.
From the field notes
Start with the phase you’re in.
Not sure which workflow? Start with the $25,000 Sprint. Already know it? The $8,000 AI Quick Win Sprint ships it. A 30-minute call is enough to tell which.