About
A small studio, deliberately.
Newtonly started because we kept being called in after the fact — to rescue AI pilots that impressed a boardroom and then quietly stalled. The pattern was almost always the same: nobody had defined what "working" meant, and nobody owned the system once the consultants left.
So we built the studio around the opposite. We take three engagements at a time. The people who scope your project are the people who write the code. And every engagement ends with your team able to run, change, and argue with the thing without us in the room.
“The interesting question was never which model. It's whether the people doing the work trust the output enough to stop double-checking it.”
How we judge our own work
Four things we hold to
Say no to the wrong project
If a rules engine and a good afternoon would solve it, we'll tell you that instead of selling a model. It costs us a project and earns the next three.
Ship narrow, ship early
One workflow, working properly, beats a platform that demos well. Scope grows from something running, never from a roadmap.
Write it down
Decisions, trade-offs, known failure modes. If your team can't read why something is the way it is, we haven't finished.
No black boxes we can't explain
Every system we hand over comes with its evaluation set, its limits, and an honest account of where it will be wrong.
HOW THE STUDIO IS PUT TOGETHER
Small team, whole problem.
An engagement is staffed by two to four people who stay with it from the first workshop to handover. There is no separate delivery team and no handoff to juniors after the pitch.
Engagement lead
Runs discovery, owns the scope document, and is the person you call when a decision needs making. Stays on the project through handover.
AI engineer
Builds the retrieval, extraction, or scoring layer and the evaluation harness that keeps it honest as models and data change.
Product engineer
Writes the software around the model: interfaces, integrations, permissions, and the parts your team touches every day.
Data engineer
Joins when the inputs need work — pipelines, schema, deduplication, and quality checks that fail loudly rather than silently.
HOW WE WORK WEEK TO WEEK
A rhythm you can plan around.
Monday
Cycle planning
We agree what the next two weeks will produce and what we are deliberately leaving out.
Daily
Shared channel
Questions, decisions, and blockers happen in writing in one place your team can read at any time.
Every second Friday
Working demo
A live walkthrough of the real system against real data, including what did not work and why.
Monthly
Measures review
Quality, cost, and response time against the baseline set in scoping, with an honest read on the trend.
WHAT WE BELIEVE ABOUT AI
A short, unfashionable position.
Most of the value in this technology will come from unglamorous work: reading documents accurately, retrieving the right paragraph, and knowing when to stop and ask a person. We would rather be good at that than early to the next demo.
Accuracy is a product decision, not a model setting
The right threshold depends on what a wrong answer costs. That is a conversation with your team, not a default we pick for you.
The demo is the easy 20%
Permissions, edge cases, failure handling, and the day the provider has an outage are where the real work sits — and where most pilots quietly die.
People should stay in the loop where judgement matters
Automation should remove the tedious part of a decision and leave the accountable part with a person who can see how it was reached.
Want to work with us?
We take on a new engagement roughly every six weeks. Tell us what you're weighing up.
Get in touch