The system is breaking: you are in the right place.It works, and you want to grow:Speedup.

Your internal tool stops breaking. Your team keeps building.

I find out why every change breaks something else, fix what matters first, and leave the knowledge in the system.

A person traces a tangled cable to its knot on a leaning tower of cubes, with a magnifying glass

Artem Kiselev

When the tool your team runs on breaks, I'm who you call. Former tech founder helping agencies and service businesses turn vibe-coded internal tools into software they can rely on.

Three steps, priced separately

Rescue is engineering work on an internal tool a business already runs on. Each step stands on its own. You can stop after the first one, and most of what the first one gives you is useful even if you never hire me again.

Health-check$1,900 · 5 business daysA written report you keep whether or not you continue. It is not a 45-minute pitch call.

Rescue$5,000 to $15,000, fixedStabilize what exists, or rebuild the part the numbers name.

Retainer$1,000 to $2,500 / monthInsurance and judgment, not support. Optional after the first step has paid for itself.

AHealth-check

$1,900 · 5 business days

Who it is for: You own a tool your operations depend on, it was built with AI tooling, and you can no longer tell whether a change is safe.
What I do: Separate the requirements your business actually has from the ones the tool accumulated.
Read the system: code, data, integrations, accounts, backups.
Rank the risks by what they cost you, not by how ugly they are.
Decide per component: fix, rebuild, or replace with something off the shelf, with cost and time for each.
What you get: A written report: prioritized problem list, a decision per component, and a cost estimate for the next step. Plus a 45-minute walkthrough by voice. The report is yours to keep, and to hand to any other engineer. See a redacted sample report.
Price and terms: $1,900, paid up front. If you continue, it counts toward the project price. It is paid on purpose: it means the people who come have a problem that is worth money, and it means I answer for the diagnosis with my time and my money, not with a free call.
Not included: No code changes, no production access, no new features. Reading and judgment only.

BRescue

Fixed price, $5,000 to $15,000

Two ways to do it. Which one applies is decided by the diagnostic, in numbers, and you see the comparison and make the call.

Default

Stabilize what exists

Stop the pain that is costing you this month. Then put practices in place on the running system, gradually, without a freeze.

When the numbers say so

Rebuild a component, or the system

With clear requirements, coding agents and practices set up from the start, a rebuild is often cheaper than untangling. It is an engineering choice, not a verdict on what you built. Your data comes with it, the old system runs until the switch, and the switch is reversible.

Timeline. Named at the end of the diagnostic, when there is something real to name. Typically two to six weeks.

Price and terms. Fixed price, not hourly. If the work takes longer than I thought, that is my problem. Half up front, half on handover.

The ceiling is a limit, not an anchor. If the work does not fit inside $15,000, the honest answer is “this is a different kind of job, and here is who does it”. Not “rewrite from scratch, $60,000, four months”.

CRetainer

$1,000 to $2,500 / month

You do not need the retainer for the system to keep working. If we part ways, everything keeps running. The knowledge lives in artifacts in your repository, not in my head.

Insurance and judgment, not support. You are buying a fast answer on the day something matters, and a second opinion before you build the next thing.

Monitoring the system and responding to incidents.
Backups, with restores actually tested.
Keeping the agent instructions current. A stale instruction is worse than none.
A fixed number of hours for changes each month.
A monthly report to you, in plain language.
A build-or-buy answer when a new need shows up.

Three-month minimum, then month to month, 30 days’ notice either way. Sold after the first step has paid for itself, not before.

Three possible outcomes of a diagnostic

No favorite among them. Which one applies is decided by numbers after the diagnostic, and the criteria are published here so you can check my reasoning.

Stabilize. The domain logic is right and the value sits in accumulated data and the edge cases nobody remembers. Keep it, make it safe to change.

Rebuild. Cleaning it up costs more than doing it again against clear requirements. Applies to one component more often than to the whole thing.

Buy it. The requirement turns out to be ordinary, and a $60-a-month subscription closes it better than any code I could write.

Decided on:

Code size and coherenceValue of the dataClarity of requirementsNumber of integrationsHow soon you need it

“Am I going to tell you to rewrite everything?”

No. A full rewrite is the rare case, not the default. If fixing is cheaper, I say so, and it is written in the report, with the numbers, before you decide anything.

What stays with you

Everything below lands in your repository, on your accounts, with your keys. If we part ways tomorrow, it keeps working and another engineer can pick it up.

Agent instructions

AGENTS.md

Hard rules
- Do not change billing math
  without a matching test.
- Never write to jobs directly.
- If a query returns 0 rows,
  check the policy first.

Format sample. Real redacted
files replace this when the
owner has given permission.

A map of the system

What is where, and why.

A runbook

Routine operations and the incidents that actually happen.

Tests on revenue paths

The scenarios your income depends on, not coverage for a number.

Backups, restored once

Configured, with a restore performed and verified.

Monitoring and alerts

Tied to business consequences.

A decision register

What was chosen, and why.

Real fragments, redacted with the owner’s permission, belong on this page. Required, not optional. The sample above shows the format until those files can be published.

Access, NDA and your data

What I need from you, and when. Production database access and admin rights are not requested before a signed agreement.

Health-check

Read access to the repository, a look at the tool as a user, and two conversations. Nothing else.

Rescue

Write access to a branch, a staging environment, and the specific integration keys the work touches. Issued by you, revocable by you.

Retainer

Monitoring and deployment access, plus whatever an incident requires. Reviewed every quarter and trimmed back to what is used.

NDA

Signed on request before any access changes hands. I have the template ready so it does not cost you a week.

Your data

I work on anonymized or test data wherever it is possible. What I do hold sits in one encrypted workspace, is listed for you, and is deleted 30 days after the engagement ends.

Ownership

Code, repository, infrastructure, keys and artifacts are entirely yours. Nothing of mine is licensed to you, no accounts are in my name, there is nothing to be locked into.

If we stop mid-way

You keep the report, every commit made so far on your own branches, and the artifacts written up to that point. I write a short handover note either way.

Where my responsibility ends

If something I touched breaks. I respond the same business day and fix it at no charge for 60 days after handover. That is a boundary, not a guarantee of perfection.

What is not covered. Parts of the system we agreed not to touch, changes made after handover by someone else, and outages at your vendors. I will help you read them; they are not mine to fix for free.

What I will not claim. Nobody can promise a system that never fails. What I can promise is that when it does, there is a runbook, a tested backup and a person who answers.

When something else is the better buy

You are going to ask your own AI this question anyway, so I would rather answer it first.

A cheap one-off fix: One thing is broken, you know which, and nothing else worries you. Enough. Take it.
An in-house developer: There is more than a year of steady work, and you can judge technical work well enough to hire and manage it.
An agency: You need volume and parallel tracks more than you need one person holding the whole picture.
Upwork, Toptal: The task is clear enough to write down, and you can check the result yourself.
This: You need it to stop breaking, you cannot verify the work yourself, and your team should still be able to build after I leave.

This page is not for you if

you are raising a round and need the product packaged for investors;
you want a $50 one-off fix;
it is a side project with no business consequence when it breaks;
you want someone who executes without the right to disagree with you;
you need a team to outsource to.

Questions I get asked

Why is the diagnostic paid?

Because five days of my reading is real work, and because a free audit is a sales call wearing a costume. Paying it also means you own the report.

Can I stop after step one?

Yes, and some people do. The report is written so another engineer, or your own team, can act on it without me.

Can my team keep building after you leave?

That is the point of the work. The instructions, tests and runbook exist so the next change your team makes with an agent does not break the last one.

Do you know my tool?

Lovable, Replit, Bolt, Base44, Glide, Softr, Airtable, n8n, Zapier, Supabase, Cursor, Retool. If yours is not on the list, say what it is. The failure patterns are mostly the same.

Why did my own AI fail to fix it?

Not because you prompted it badly. The agent starts every session with no memory of the system and no way to know what it does not know.

How much of your work is done by agents?

A lot of the typing. None of the decisions. I read every change before it reaches your repository, and I answer for it.

What if you disappear?

Everything is in your repository and on your accounts, written for someone who has never met me. That is the honest answer to the bus question.

Do you need production access?

Not for the diagnostic, and not at all before a signed agreement. For the work itself, only the specific access the change requires.

How do you handle client data?

Anonymized or test data where the work allows it. What I hold is listed for you and deleted 30 days after we finish.

How fast can you start?

I work with three clients at a time, so usually within a week or two. If it is longer, I say the date instead of holding you.

Who actually does the work?

Me. Diagnostic to operations, one person, nothing lost in a handoff. There is no team to introduce you to.

Can I run this page past ChatGPT first?

Please do. Everything here is plain text on purpose. Prices, scope, limits, so it can be checked without talking to me.

Start with the diagnostic.

Five questions about your tool, then payment. I read your answers myself and reply within one business day with a start date. In five business days you have the report, and no obligation to go further.

Start the diagnostic me@artkiselev.dev

Write at 11:40pm on a Sunday if that is when you have time. I will answer, and it will be me answering.

Artem Kiselev. When the tool your team runs on breaks, I'm who you call. Former tech founder helping agencies and service businesses turn vibe-coded internal tools into software they can rely on. More about me