Adrian Rusan · Senior full-stack engineer, 10+ years
I ship your backlog with AI agents and review every PR myself.
Security first: injection, auth, secrets and access control, checked line by line before a PR reaches you. Expect 2–3x the throughput, not 100x.
If the first reviewed PR isn't merge-ready to your standard, you don't pay for it.
Review log
- #112Mar 2026 · own dashboard · A job ID from the URL went into a shell command. Fixed before merge.
- #111Mar 2026 · own dashboard · A search query from the request went into a shell command. Fixed before merge.
- #119Mar 2026 · own dashboard · A notes path built from a route param, with no containment. Fixed before merge.
- #109Mar 2026 · own dashboard · Database values in a shell string. I missed it; my October re-audit found it and 4 more.
- #39Oct 2026 · this site · The case-study placeholder check ran only in tests. Moved into the build.
From my own repos, not client work. The 30-PR case study lists every catch and all 5 misses.
Agents write code fast. Someone has to decide what merges.
Agents can write most of a backlog, and they write injection bugs too. On my own 30-PR run, with no tests and no CI, review was the only gate. Agents without review ship bugs faster.
How it works: agents ship, I review
1 · Ship
The agents write the volume
Several Claude Code agents run in parallel git worktrees, one ticket each, following your CLAUDE.md. That's where the 2–3x comes from. For scale: 632 merged PRs in 32 active days (11 Sep–13 Oct 2026), client and own work together, counted from my engineering ledger.
2 · Verify
I review every PR
Every agent PR gets a line-by-line adversarial review before it reaches you. Security comes first: injection, auth, secrets and access control. Then correctness and scope. I assume the agent got it wrong until the code shows otherwise.
See a sample report →
Case studies
- Own project
3shell-injection paths from request input caught in review. The study lists the 5 security bugs it missed.
30 Agent PRs, One Reviewer: What Review Caught, and the 5 Security Bugs It Missed
Read it → - Client work (NDA)
0the true count behind a landing-page line labelled "real number". The gate keeps lines like it off the page.
The Honesty Gate: Stopping Agents From Publishing Invented Facts
Read it → - Own setup
57%lower cost per merged PR in one month, then it rose 42% by early October
Cost per Merged PR: What 3 Months of My Claude Code Ledger Show
Read it →
What people I've worked with say
“Meticulous, smart and proactive. A true craftsman. Rock solid work.”
Bogdan Tufeanu, CEO, DentStore “Adrian was reliable at every step. He ran two product builds in parallel, kept us informed every time, and told us up front when an estimate was too optimistic. He takes problems over and drives them forward without anyone having to push.”
Alexandra Stroe, Co-founder & COO, Bono Fintech
Offers and prices
Every offer has a published starting price, and the final price is fixed before work starts. Move up a tier within 30 days and what you already paid counts toward it.Prices exclude VAT; VAT is added where it applies.
Plan
from €200
A written build plan, or ranked findings with the top fix shipped
Architecture Session · AI Code Audit
Set up
from €3,000
Your team's rules file and agent setup, running on your repo
Team workshop
Build
from €3,000
One automation or agent setup, built and handed over
AI Automation · Agent Fleet Setup
Ship
from €1,000
Your backlog, shipped and reviewed
Reviewed PR · Delivery Sprint · Retainer
The review layer is one senior human, so I run one sprint or retainer at a time.
Who this is for
A good fit if
- You lead engineering, founded the company or own delivery at a funded startup or scaleup, especially where security review is non-negotiable: fintech, health, payments, anything handling money or PII.
- You're drowning in backlog and can't hire your way out fast enough.
- You want agent-speed throughput but need a senior human accountable for what merges.
Not a fit if
- You're shopping purely on hourly rate. The review layer is the value you'd be pricing out.
- You want to hear "100x" and don't care who reviews the code.
- You need a staff-aug seat with a badge and standups. I sell reviewed PRs at a fixed price.
Who reviews your code
I'm Adrian Rusan, a senior full-stack engineer with 10+ years in production. I work through my own Romanian-registered business (PFA), most recently as founding engineer for Bono Fintech. Before that, live graphics for PGL Major streams with 2M+ concurrent viewers, 300+ survey projects at Ipsos for global brands, and ad dashboards at GoSocial. The agents run on my own harness: 20+ skills and 20+ hooks that scan for secrets, block destructive commands and refuse a merge when a check can't run. I still read every PR myself.
Questions CTOs ask
How is the quality any good if AI agents wrote it?
The agent writes a draft. I decide what merges. Every PR goes through a line-by-line senior review that assumes the agent got it wrong until it's proven right. Agents give you 2–3x throughput, not 100x, and they get specific things wrong. The review is where those get caught, and every sprint ships you a log of what got caught in your codebase.
Unsupervised agents ship vulnerabilities. How do you handle security?
Correct, they do. Veracode's July 2026 test found a known vulnerability in about 44% of AI code-generation tasks. Security is the first pass of every review: injection, auth, secrets, access control. On my own 30-PR run, a repo with no test suite, the review caught 3 shell-injection paths from request input before merge. The guarantee is in writing: an in-scope vulnerability found in a PR I marked reviewed gets fixed free.
Who owns the code? Will you sign an NDA?
You own everything from the first commit: your repo, your branches, your IP. I sign your NDA or DPA before I get repo access, work happens in your repository under your access controls, and access is revoked at handoff unless you're on a retainer.
Who pays for the LLM and API usage?
I do. LLM/API usage is included in the fixed price for sprints and reviewed PRs. No token bill arrives on top. Agent Fleet Setup is the exception: it runs on your team's own Claude Code licences, because the harness stays with you.
Why not an agency, more engineers, or an AI review bot?
An agency sells a team's hours with a markup, and new headcount takes months to hire and ramp. An AI review bot is cheap and worth running, but it can't be held accountable for a merge. Here one senior engineer runs several Claude Code agents, and every PR still gets my line-by-line review. The first reviewed PR lands within days.
What happens after I book a call?
Book a slot and add one line on what you want built or shipped. That line is all the prep. We look at your backlog, your repo or your AI question. I tell you which tier fits and what it costs, or that it isn't a fit. Within one working day: the scope, the fixed price and a start date, by email. The decision is yours, and I don't follow up unless you ask.