Turn one painful business process into software that just works.
I map the process, decide what should actually be automated, and build the smallest reliable system that fixes it. It starts with a fixed-price audit, so you know what to build, why, and what it will cost before you commit.
Or describe your process first →A process is costing you hours every week
For growing companies with a small tech team, or none at all, where important work still moves by hand between inboxes, spreadsheets and internal tools.
The goal is a concrete change. Your finance team stops copying data between three systems. Your operations team stops depending on the spreadsheet nobody dares to touch. Your support engineers start from a diagnosis instead of raw logs.
Workflow first. AI second.
Most business problems don't need an autonomous agent. They need a well-designed system, with AI in the few places where it adds value.
Start with the workflow
Most steps in a business process should be plain, tested code. AI goes where the input is genuinely fuzzy, behind validation, with a human where a mistake is expensive.
Small before big
The smallest system that removes the pain ships first. The rest goes on a list, and the design leaves room for it.
Production, not demos
Tests, logging, deployment and docs are part of the build, not an afterthought. You own the code and can run it without me.
Audit. Build. Ship.
Three steps, each with a fixed scope and a fixed price. You can stop after any of them and keep what you have.
Audit
Map one process, decide what to automate and what not, and design the smallest system that fixes it. Ends in a fixed-price proposal.
Build
Your team builds it with the plan, we build it together, or I build it. Tested, reviewed code from week one.
Ship & run
Deployed on your infrastructure with CI, logging and monitoring sized to the system, and docs. Optional retainer for support and the next workflow.
The Scope First Audit
One process. One week. One clear plan.
At the end of the week you'll know:
- What to automate: the process mapped step by step, and for each step whether it needs plain code, an AI model or a human. Including where software shouldn't go.
- What to build: the smallest system that solves the problem, with its architecture, data model and failure modes.
- What it will take: a fixed-price proposal with scope, timeline and price, ready to sign or to hand to another engineer.
- How to ship it: deployment, testing, monitoring, and who owns what once it's live.
Delivered as a written report and a walkthrough with your stakeholders and engineers. You keep the work, whether you hire me or not.
Why paid? Because the audit is the work. A free pitch gets you a guess. A paid audit gets you a plan you can act on, with or without me.
A manual process doesn't have to cost millions to be worth fixing. Three people spending five hours a week on it, at €40 an hour, is roughly €30,000 a year. The audit tells you whether the opportunity is big enough to justify building.
Build it your way
The audit recommends one of three paths. Each is quoted at a fixed price in the proposal.
You build it
Your engineers take the plan and run with it. I step back. If you want an independent review when you are done, I already know the problem, so it is fast.
Build it with me
Your developers build, I design the architecture, pair on the hard parts and review every pull request. The knowledge stays in your team.
I build it
I build the system using AI agents where they add leverage, with deterministic code, tests and human review around anything that matters. You get the code, the tests and the docs.
Ship it, then keep it running
A demo isn't a system. It's done when your people rely on it, and it fails loudly instead of quietly.
- CI with formatting, linting, type checks and tests on every change
- Deployment on your infrastructure, with backups, and logging and monitoring sized to the system
- Docs and a handoff session, so your team can operate and extend it
- Optional retainer, scoped in the proposal: fixes, improvements, and the next process to automate
Example: support ticket triage
An illustrative example, based on the kind of process companies bring. Not a client case.
The situation
An equipment supplier with a six-person support team. Tickets arrive by email with log files attached. Engineers spend the first half hour of every ticket reading logs to work out which known problem it is.
What they asked for
"An AI agent that answers our support tickets."
What the audit found
Most tickets match a known log signature: that's pattern matching, plain code, no AI. Where the free-text description is genuinely ambiguous, a model can classify it into a fixed set of categories. Replies to customers stay with an engineer.
- Tickets come in.Email and attached logs are parsed into a structured ticket.
- Known signature?Plain code matches the logs and produces a diagnosis plus a draft reply.
- Unknown?A model classifies the description into a fixed category, and the ticket lands in an exception queue.
- An engineer approves.Every reply is checked by a human, and each newly solved problem becomes a signature rule.
The proposal: an ingest and diagnostics pipeline with a review queue, a 6-week fixed-price build. Engineers start every ticket from a diagnosis instead of raw logs, and every new problem they solve becomes a rule. No autonomous agent needed.
Replacing my paid bookkeeping tool
My own business, so I was the client. Same method, same constraints.
The situation
As a freelancer in Spain I invoice clients, collect supplier invoices in several currencies, and hand a monthly close to my accountant. I paid for a bookkeeping tool that did the job, but it was slow to work with and I couldn't get the reports I needed out of it.
What I wanted
Something fast that fits how I already work, from the command line, and a monthly report my accountant can check without asking me.
What the audit found
Only one step needs AI: reading supplier PDFs, which come in every layout. Numbering, VAT, withholding, currency conversion and monthly totals are rules and arithmetic, so plain code with tests.
- Drop the PDFs.Supplier invoices go into one folder, any filename.
- AI reads, a human approves.A model extracts supplier, date, currency, net and VAT. I approve, edit or skip each row before anything is written.
- Plain code does the rest.Foreign amounts are converted at the ECB rate for the invoice date, totals are computed, and the ledger is validated to the cent.
- The accountant gets a traceable file.The monthly reconciliation workbook is one command. Every figure is a formula pointing back to its source row.
The result: the subscription replaced by about 2,000 lines of tested Python. The first version took two days; the weeks after were refinements from real use. Closing the month is now a single command, invoices can't be overwritten, and the data lives in plain files I own.
Bob Belderbos
I've built software for 20 years, most of it automation: 13 years on critical production environments at Sun Microsystems and Oracle (2007 to 2020), then co-founding Pybites and running its coding platform in production for years.
I've also coached 150+ developers through thousands of code reviews, which taught me to spot design problems early and to transfer ownership instead of creating dependency.
More about me →Before you book
Is my process a good candidate?
If it runs every week, takes 10+ hours of someone's time, and follows rules people could write down, almost certainly. The fit call tells you in 20 minutes.
What if we don't go ahead after the audit?
You keep everything. The plan is written so any competent engineer can build from it.
What happens on the fit call?
20 minutes on your process: how often it runs, who does it, where it hurts. If it's a good candidate, I send the audit scope. If it isn't, I'll tell you.
Who owns the code?
You do. All code and infrastructure built during the engagement belongs to your company.
Do we need to use AI?
No. The audit decides where AI adds value. Most of a reliable business system is ordinary code, rules and tests. AI goes where it improves the outcome, not because every workflow needs an agent.
What will the build cost?
It depends on the process, which is why I don't quote before the audit. The audit ends in a fixed-price proposal based on the real scope, so you know the number before you commit to anything.
When is this not a good fit?
If you already know exactly what you want built and just need developers to implement a spec. My value is in deciding what to build; a development agency is the better choice for pure implementation.
Do we need our own developers?
No. Build from the audit with your existing team, build it with me, or have me build it. The audit is yours either way.
Have a process worth automating?
Book a 20-minute fit call. Bring the process, not a spec.
Rather write it down? Describe your process →
Prefer email? Drop me a line →
Not ready to talk? Start with Scope First →