Governance
Vibe-coded prototype
Not in place.
Production system
Owned from day one.
We decide which AI initiatives can reach production. Then we build them.
For teams of 50 to 500 people, the bottleneck is rarely the model. It is the choice, the workflow, the owner, and the handover.
The production gap02
The gap between pilot and production.
A pilot runs in a week. A production system runs for years, under real users and real rules. That is the gap. Five reasons pilots stay stuck:
Why pilots stay stuck
- The wrong use-case: AI solves something that was never the problem.
- A pilot is not a production system. Demo code can't hold 500 real users.
- No workflow integration, so the thing sits next to the work instead of inside it.
- Governance comes too late. Compliance blocks at launch, not before.
- ROI is never measured in operation, so no one knows if it paid off.
Compliance
Vibe-coded prototype
After the fact, when it breaks.
Production system
Up front, in the design.
Maintenance
Vibe-coded prototype
No one, it's a demo.
Production system
A team that owns it.
Key-person
Vibe-coded prototype
One builder leaves, it falls over.
Production system
The knowledge lives in the team.
In numbers
- 12 years
- In data and AI
- 35+
- AI implementations witnessed close-up
- 95%
- AI pilots without measurable ROI
- max 4 weeks
- Per production sprint
Engagements03
Three engagements. From scan to production.
Every engagement starts with a written brief and ends with something you can run without us. No discovery phase that drags for months. No retainer without an end date.
Scan
AI ROI / Production-Readiness Scan
Two weeks. We look at your pilots and tell you which to kill, scale, or rebuild. Fixed fee.
- Inventory of your pilots
- ROI-potential score per pilot
- Production-risk score per pilot
- Verdict: kill, scale, or rework
- Top 1 to 3 candidates for production
- 90-day plan
Pilot-to-Production
One workflow live
Max four weeks. We ship one workflow to production, paired with one of your engineers so the knowledge stays in-house.
- One workflow live in production
- Governance and compliance built in
- One of your engineers trained alongside
- Documentation and metrics
Operating Layer
Managed evolution
A retainer with a goal and an end date, never open-ended. Governance, maintenance, and the operating model underneath.
- Governance on production
- Maintenance and monitoring
- The operating model your team runs
- A pre-agreed end date
How we work04
Seven steps. Ends with you.
We make the architecture visible while we build. No black box. No handover on the final Friday.
We map your pilots.
We assess them on ROI, feasibility, risk, and production-readiness.
We recommend: kill, scale, or rework. Sometimes our most honest advice is: stop this pilot.
We pick one workflow.
We build and harden it with a small senior team.
We put the operating layer underneath.
We hand it over.
What we stand for05
Three rules. No exceptions.
Lessons from projects that worked and projects that stayed stuck.
Advice may end with no.
Most of what gets pitched as AI is a SQL query, a rules engine, or a bad idea. We tell you which yours is, and we say no when no is the answer.
Depth over speed.
A demo is not a system. We build the layer underneath: governance, compliance, and maintenance, so the system can continue after the first release.
You get more independent, not more dependent.
We pair with your own engineers from day one. When we leave, your team runs the system itself. We build ourselves out of a job.
No lock-in. We leave you with documentation, an operating model, metrics, and ownership.
What we don't do06
What we don't do.
- No standalone training as the end product.
- No prompt workshops as the main product.
- No advice without a path to production.
- No junior staffing under an AI label.
- No fragile prototypes that leave you dependent.
Writing07
Notes from the field.
- 04 / 2026
- 04 / 2026
What 35+ AI implementations taught us.email
- 03 / 2026
- 03 / 2026
- 02 / 2026
Who's behind this08
Thijs Bongertman.
12 years in data and AI. KPN, Reaal, Eneco, then Hamburg as Head of Data at Free Now. Five years in Germany, partly at a startup where I built the data function. When that startup collapsed, I started out on my own. Master's in Tilburg. Now in Rotterdam.
At that startup we bet €300k of the company budget on a GenAI product. It shipped. It didn't work. That lesson is the foundation I build on now: I check every recommendation against one question, would I build this myself?
That's how we pick people too. No passengers. No ticket-takers.
Get in touch
Put one AI initiative on the table. We will say what should happen to it.
In a working session we triage your pilots: what can move to production, what should be killed, what still needs work.
honest.ai