Production experience meets frontier AI.
What any guided expert from any field can achieve with AI is vastly bigger than was ever possible. AIntelligent Technologies is a new kind of software engineering company. Seven years of production experience in building, scaling, and shipping real software, combined with frontier AI and bespoke governance tooling. We build production-grade software and AI products, and provide AI strategy advisory.
The Garden Birbs reveal
With sound. Press play.
How We Gave Stateless Hooks Memory: The Sentinel Pattern
Lifecycle hooks are stateless, but agent governance often depends on something that happened one or two events earlier. Here is the pattern we built to bridge that gap.
Read the article: How We Gave Stateless Hooks Memory: The Sentinel PatternAn evaluation starts from the code. A build starts from what you need.
UK-based.
We evaluate software from the code outward and build custom projects around your requirements.
Your question decides where we look
An evaluation can cover source and tests, runtime and deployment, browser journeys, crawler responses and authorised security checks. Your question and the access available decide which apply.
Findings come with the evidence behind them
Agree what the written record needs to answer, including what was examined, what was found, how it was checked, and what could not be observed.
A finding and a change are separate work
An evaluation reports what it finds. A change to the product, or a custom build, is its own piece of work.
How we work
We check whether a test can fail
Coverage tells you which lines of code ran during the tests. It does not tell you whether anything would have complained if those lines were wrong. That second question is the one worth asking, and it is the one we answer.
Rules that block, not rules that remind
When a check costs nothing to run, it becomes worth enforcing things nobody would accept being nagged about by a colleague. Those turn out to be the ones that catch real bugs.
A count that is only allowed to fall
Switch on a strict new rule and it fails in a thousand places at once, so somebody turns it off by Friday. Instead we write down today's number and only allow it to go down. One of ours went from 177 to 142 in 28 days, counted on 2026-09-07.
How you can check us
The standard lives in the written rules, not in one person
If the quality of the work depends on who is doing it, what you are really buying is that person. On one of our builds, 24 of 27 runs used the cheapest model tier, and two full-tree audits found zero layering violations.
The documentation gets written because it has to be
On a team, the things people just know are free to carry around and expensive to write down, so they never get written down. They leave when the person leaves. We had no such option: anything not written down simply does not get used. Our standards exist because the alternative was guesswork.
No number we have not measured
We ship what works and say plainly what is still coming. No inflated claims, no invented testimonials. If we have not measured something, it does not appear here.
From the blog
Deep dives into engineering discipline, development velocity, testing confidence, and how we build production AI products without letting standards slip.
Building ESLint Rules to Prevent Tests That Lie
How we turned our false-confidence taxonomy into three AST-level ESLint rules that catch deceptive test patterns before they reach CI.
60% of Our Tests Had Zero Signal: How We Discovered False Confidence
A systematic audit of 656 test files revealed that most of our test suite was providing the illusion of coverage while catching nothing.
Four products. Four design systems. One method.
A method that produces four versions of the same template has not produced anything. Each band below is that product's own design system, quoted inside this page rather than translated into ours — its ground, its type, its primitives, its voice.
Your business is in eleven places and its answers should be in one. Business OS reads the tools you already pay for, keeps what it learns, and works on a schedule — then tells you what needs you. It is also the instrument this practice is run from, and the system most of the standards on this page were worked out against.
Business OS · closed alpha
Your business is in eleven places. Its answers should be in one.
Business OS reads the tools you already pay for, keeps what it learns, and tells you what needs you — then does something about it. Click anything below.
The vault
Your business, written down
GOALS · CLIENTS · DECISIONS
COMMITMENTS · OFFERS · CONTRACTS
The agent
Relevant context, every time
YOUR MODEL · YOUR JUDGEMENT
And it runs itself
On a schedule, or when you ask
Click anything. Every source shows what it produces; every output shows what feeds it. Nothing here is a category we invented — each is something you already use, or something you already wish somebody did.
Press Escape to clear.Tap it again to clear.
A local-first garden companion, and the deliberate opposite of the flagship: paper instead of terminal, characters instead of a checklist, and nothing to win. Same engineer, same year, same rules.
- jack. jackdaw. pale eyes, strong opinions.
- no life list. no leaderboard. nobody is winning.
- i sit on the chimney pot. that is the whole feature.
- it remembers which jackdaw i am. that took some doing.
- peanuts are off in fledgling season. sorry.
hey, jack’s up
garden birbs is for the people who watch the same four birds every morning and have named all of them. jack is a jackdaw with pale eyes and opinions, and he nests in the left chimney pot. the app knows which jackdaw he is.
no species counts, no rarity badges, no follower graph, no streak to lose. a garden is a place, the regulars are characters, and the whole thing is drawn as stickers on paper.
- local-first
- no life list
- no leaderboards
- care without guilt
a minute at the window, written down. with sound.
NDT Platform
An alert network for UK non-destructive-testing work: contractors keep a certification matrix, employers publish a vacancy, and the platform tells the contractors who qualify. Its language is engineering drawing on technical paper.
NDT Platform — sheet 1
Matched by certificate,
not by keyword
Inspection work runs on documents: procedures, drawings, inspection reports, cert cards. So the product presents itself as the best-drafted technical file in the industry — drafting grid, dimension lines, figure captions, mono facts, and international orange as the one accent, the colour of survey markers and site equipment everywhere.
A contractor keeps a structured certification matrix and their availability. An employer pays to publish a vacancy. The platform matches the two and alerts the contractors who qualify, rather than leaving either side to read a job board.
- SECTOR
- UK NDT & INSPECTION
- CONTRACTOR
- CERT MATRIX + AVAILABILITY
- EMPLOYER
- PAID VACANCY, PUBLISHED
- MATCH
- ALERT TO WHOEVER QUALIFIES
- PUBLIC PAGES
- SERVER-RENDERED
- STATE
- PRE-LAUNCH
Rotten Hand
A realtime party card game for up to sixteen people, and the furthest thing here from a business tool. Its language is riso print: craft paper, spot inks, a display serif set large because the writing is the product. The engineering underneath is the same as everything else on this page — a client is only ever sent card text it is allowed to see, and a test plants a marked secret to prove it.
Rotten Hand — the table
Fill in the blank.
Funniest card wins.
Up to sixteen people, one prompt, and a hand of cards nobody else can see. A comedy word game whose cards look screen-printed rather than generated — craft paper, riso spot inks, a display serif set large because the writing is the product.
The interesting part is underneath. A client is only ever sent card text it is allowed to see, so a player who opens the network tab learns nothing: the rule is enforced by one projection helper and a wire-level test that plants a marked secret and asserts it never reaches anyone else. Every card is written to a published house standard and checked against it on every change.
The sprint slipped because of ______.
Pick 1a whiteboard nobody can reach
one more quick thought
More to come
The same practice, pointed at smaller things. Open-source tooling, agent skills, and a few side projects that exist because they were worth building.
Open-source tooling
Parts of the engineering practice on this page are being cut loose as standalone tools — the type-aware navigation, the gate design, the false-confidence checks.
Agent skills
The reusable workflows our own agents run, packaged so other teams can run them too.
Versecraft
A free educational project. No account, no price, nothing to buy.
Philosophycraft
The same idea pointed at a different subject. Further out.
Lucid, a RimWorld mod
Headed for the Steam Workshop.
Four ways this usually starts
Taking over software, considering a rebuild, or investigating a recurring problem? Bring the decision your team needs to make and what you already know about the product.
Find out what your code actually does
A software evaluation can examine source and contracts, tests and seams, runtime and deployment, browser journeys, crawler responses and authorised security checks. Agree what the written record needs to answer: what was examined, findings and their evidence, proposed changes and questions still open. A change to the product, or a custom build, is its own piece of work. If the product cannot leave its environment, talk to us about what can be examined in place.
Build what you need
We build custom projects, including websites. We write decisions down and run checks automatically as part of the build. Tell us what the software needs to do, who will use it and what it needs to connect to.
Get a struggling project back on track
Something that has stopped shipping, or ships and breaks. We work out why before proposing what to do about it, and say plainly if the answer is that you do not need us.
Technical and AI advisory
Where AI helps, where it does not, and what running it involves either way. Tell us what you would want AI to do, who would use it and what it needs to connect to.
Get in touch
Whether you have a question about our products or want to discuss AI advisory, we would love to hear from you.
Details
- contact@aintelligenttech.com
- +44 7533 172442
- Derbyshire, United Kingdom