Agent fleet in daily use
Task automation, drafting, monitoring, reporting and handoff — built, operated and owned end to end.
Riga, Latvia · UTC+2/+3 · fully remote
AI project manager
Twenty years getting people who do not report to me to change a process that does not work — in a bank, in a municipal company I directed, and for seven years in national public administration. Since 2025 I build and run my own AI agent systems in daily production.
Ask this page
You have a question about my background. Instead of scrolling for it, ask. This box runs on a model wired to my CV and nothing else — the same kind of system I build for a living. It will refuse anything outside that CV.
Generated live from his CV. Not a chat with Maksims — for anything the CV does not cover, write to maksims.burcenko@gmail.com.
In numbers
These are volume figures, and they are true. I do not publish percentage improvements for my AI work, because I do not have measured before-and-after numbers I am willing to state as fact.
Summary
I am a project manager. For twenty years my job has been to take a process that does not work, find out why, and get people who do not report to me to change it. Since 2025 I have built and operated my own fleet of AI agents in daily production use. That is not a hobby beside the work; it is the same work with better tools, and it is the reason I now start from the process rather than the software.
AI adoption does not fail because the tool is bad. It fails because the staff do not use it.
Current practice · 2025 — present
Independent practice, run solo and unsupervised — which is also the honest proof that I work remotely without someone standing over me.
Task automation, drafting, monitoring, reporting and handoff — built, operated and owned end to end.
I author and iterate the instructions that define how agents behave, and rewrite them when real usage proves the instruction was wrong.
Claude and Gemini. Model choice per task, weighed against cost and output quality — including version changes and failure scenarios, not the assumption that one model stays available forever.
Local database, local memory layer. Every external call is a deliberate choice, not a default.
Connecting agents to the systems they need — browser control, databases, scheduled jobs — and troubleshooting when the connection is the problem.
Prompt and SOP engineering, so a workflow can be repeated by someone who did not build it.
Governance
An agent without a boundary is not an efficiency; it is an incident waiting for a date. The boundary is written first, not after the first mistake.
Allowed outright
Tools the agent may use on its own — reading, searching, drafting, local processing. Reversible work inside my own systems.
Requires human approval
Anything that leaves the machine, costs money, or cannot be undone — publishing, sending, external calls, deletions. The agent prepares; a person releases.
Forbidden
Credentials, financial transactions, destructive operations, and any action a person did not ask for. No exception is granted at runtime by the agent itself.
A solution that worked last month can quietly break after the next model release. That is why checks get re-run after a model, prompt or data change — silent breakage is the normal failure mode, not the rare one.
How I got here
I started my career in a bank, so the demands of a regulated environment are not new to me — including a calm expectation that AI adoption there moves slower than it does in a demo.
Experience
Competition Council of Latvia, Riga
2021 — 2025
National Health Service of Latvia, Riga
2018 — 2019
Templine (Kettering); Rigid Containers Ltd (Desborough), United Kingdom
2016 — 2018
SIA Psaiho.lv, Riga
2011 — 2015
PSIA Jurmalas Attistibas Projekti, Jurmala
2010 — 2014
SIA Sintegra, Microsoft Gold Partner, Riga
2008 — 2009
AS SEB banka, Riga
2005 — 2008
2019–2021 is a deliberate gap: parental leave after my elder daughter was born. I state it rather than let a reader guess.
Working across borders
Four years remote inside an institution
My entire 2021–2025 role at the Competition Council was worked remotely. It began because the pandemic made on-site work impossible and simply stayed that way. Every line in my delivery record from those years — the programmes, the budget reporting, the cross-agency working group — was produced remotely.
Remote and unsupervised
Since 2025 I have run an independent practice solo: my own priorities, my own deadlines, my own systems. No one checks whether I started.
English as working language
Lived and worked in three countries outside Latvia — United States (2003), New Zealand (2010), United Kingdom (2016–2018). ILM Level 3 taught and assessed in English.
People who don't report to me
Founded and chaired a cross-agency working group of volunteers with no reporting line to me; seven years coordinating ministries, courts, prosecution and anti-corruption bodies.
Contract and self-employment
Company director and board member; comfortable operating as a B2B contractor with my own invoicing and reporting.
Time-zone overlap
Riga, UTC+2/+3. An afternoon-to-evening shift overlapping US Eastern hours is workable.
Software product organisations
To learn. My operational depth is public-sector and services delivery, not a software product company. I would trace the process end to end before telling anyone how it should run.
Education & certification