We build AI the way software that matters has always been built
01
The contradiction we started from
In 2025, Stack Overflow asked tens of thousands of developers what they thought of AI. Eighty-four percent said they use it or plan to. Three point one percent said they highly trust what it produces. Forty-six percent said they actively distrust it.
Source: Stack Overflow Developer Survey 2025
Read those together and you find something odd: an entire industry adopting a technology it doesn't believe. Not refusing it — adopting it, at scale, while quietly assuming it will be wrong.
That gap is where we work.
02
What actually happens in companies
The pattern barely varies. Someone builds an AI demo. It's genuinely impressive. It goes to production and meets what demos never meet — the malformed record, the API that rate-limits harder than its documentation admits, the edge case that only appears at month-end, the user who does something nobody anticipated.
It doesn't fail loudly. It fails quietly, and someone finds out three weeks later.
Meanwhile the cost of generating software has collapsed. Tools that write applications from a prompt went from roughly $100M to $500M in annual revenue in under a year.
Source: Sacra, 2025
Generating an application is nearly free. Generating one you can run a business on has not become one bit easier.
03
Why we exist
This business runs on systems its founder built himself. Its own CRM pipeline — four workflows, 64 nodes — went through three independent audits in nine days, and those audits caught six classes of defect before any of them touched a live record. The pattern those audits were built to catch is the one this company exists against: software that demos beautifully and then quietly rots in production.
We started Orbav on a position that's close to heresy in an AI company: AI shouldn't be trusted with the things that matter — and that's fine, as long as someone is.
So we use AI aggressively. It writes code, drafts tests, explores approaches, and compresses work that used to take weeks. Then a senior engineer decides what's actually correct, what breaks under load, what happens when the model is wrong, and whether the thing should have been built at all.
That engineer's name is on the engagement. You'll know it before you sign.
04
What we believe
- Accountability is the scarce resource
- Model access is universal. Judgement, and the willingness to be answerable, are not.
- The standard is production, not demonstration
- Evaluations, guardrails, error handling, monitoring and documented handover aren’t premium add-ons. They’re the difference between a system and a demo.
- The job isn’t finished when it ships
- A system that quietly stops working in month four was never delivered.
- Saying so plainly is an advantage
- We publish what we start at, we say what we won’t do, and we tell clients when they shouldn’t build something — including when it costs us the work.
05
How accountability actually works here
Every engagement has one named senior engineer. Before you sign, you know who they are.
They personally sign off on:
- Architecture and integration design
- Evaluation design and guardrail specification
- The security and data-handling surface
- The handover documentation
- The decision to go live
They're on the kickoff call, the weekly checkpoints, and any incident call. If they change, that's a conversation with you, not an internal reassignment.
06
The team
Ahsan Saleem
Founder & Principal Engineer
He built the automation and CRM systems this business runs on — documented in Work.
We're not hiring right now. When we are, it'll be here first.
07
Where this goes
In ten years, “built with AI” should no longer be a disclaimer — because the standard for accountable AI engineering was set by us, and eventually sold as a product. Until then it's done by hand: every engagement, every reconciliation, every go-live decision.
How we handle your data, where we're based, and what we don't hold yet — see the Trust page