Ionuț Vizitiu, Bucharest
I run my own company through a team of AI agents. They write and test code, read documents, build websites, track my tools and brief me each morning. I approve every step that matters. I can build the same for your company.
350+
tasks finished by agents on my task board
3 h 25 min
from idea to a tested app, built overnight
60
systems on one live map, backups restore-tested every week
94
tools evaluated and tracked for cost and use
I spent over ten years building large Angular platforms for enterprise teams, most recently as a lead engineer and engineering manager. Then I tried something different. I set out to run a one-person company as if it had a staff, and made the staff AI agents.
It was not a demo. The agents write and test code, research markets, build websites, watch my tools and budgets, and brief me every morning. Some of it failed. Each failure became a rule or an automatic check, so the same thing would not break twice.
What came out is a working system, not a slide deck. It is also the clearest way I know to show a company what agents can do today, and what they still cannot.
What I built
Agents plan, build, test and review code in stages. Automatic checks match each change to its plan and ask for proof. Nothing ships until a person approves it.
Proof
One app went from idea to a tested build in 3 h 25 min, with 319 unit tests and 54 end-to-end tests.
For your company
Your team gets AI speed without losing review, the audit trail or control of what goes live.
What I built
Agents read the documents, pull out what matters and draft the reports. A fixed review turns each case into a clear decision. A person approves the decision and everything that goes out.
Proof
An audit report where every number traces back to its source line, and business reviews whose signed approval goes stale the moment the page changes.
For your company
Invoices, intake forms, reports and audits get swept automatically. The judgement stays with your people.
What I built
A chief-of-staff agent plans the day, watches the task board and hands work to specialist agents for code, websites, SEO and research. I reach them from chat on my phone, behind a login.
Proof
12 agent profiles and 26 scheduled jobs. Every outside action waits as a draft until I approve it.
For your company
Start with one agent for one process. Add specialists only when they earn their place.
What I built
Every service, schedule and backup sits on one live map. Agents work in locked sandboxes with an allowlisted network. Backups go offsite encrypted, and a restore test runs every Sunday.
Proof
60 systems and 70 data flows on one map, kept honest by tests.
For your company
You see what runs where, and you can trust it still works next month.
What I built
One private place for notes, decisions, lessons and tool evaluations. People search it, agents read it, and nothing that changes the rules goes in until a person confirms it.
Proof
94 tools evaluated and tracked for cost and use.
For your company
Decisions stop disappearing into chat threads and inboxes.
What I built
Each agent remembers past work within a strict size limit. Once a week the lessons get reviewed: repeats merge, noise drops out, and real mistakes become permanent checks.
Proof
The last review had 127 candidate lessons. One was worth turning into a rule. The filter matters more than the memory.
For your company
Agents stop repeating the same mistakes, without quietly rewriting how your company works.
Agents draft. People decide anything that leaves the company, costs money or changes a rule.
Every change carries evidence: a test run, a screenshot, a trace back to the plan.
A script before a small model, a small model before a large one. Many rules need no AI at all.
Your servers, your cloud account, or a machine in your office.
If something was not measured, I say so.
30 minutes. You describe the process that eats your team's week. I tell you plainly whether agents fit.
One process, built with your team, with the human gate and the proof in place from day one.
We measure the time it saves and fix what breaks. The next process comes only after the first one pays off.
I'm Ionuț Vizitiu, a software engineer in Bucharest. For over ten years I built enterprise Angular platforms in regulated, complex environments, moving from engineer to lead engineer and engineering manager. Today I build AI agent systems under BuildToLast, my company.
LinkedInTell me about one process. I'll reply within 1-2 business days with an honest view of what an agent could take on, and what it should not.