Marcin Kasiak
- Licensed professional engineer
- Doctorate in structural engineering
- 20+ years in AEC practice
- Author & researcher, AECO.digital
PhDIWEPMPPE
About this work
Independent research on AI accountability in licensed practice — by a practicing engineer.
The mission
Evidence that a licensed professional stayed in charge.
An audit-grade record built to be read by E&O carriers, license boards, and opposing counsel. Not a policy document. Not a consultant report. Not a generic compliance tool. A responsibility-chain record, written as the work is done, that survives multi-year scrutiny — and never claims to have made the engineering judgment for you.
The vision·2030
By 2030, every sealed deliverable that AI touched carries a record of who stayed in responsible charge of it.
The same way every structural calculation today carries a code edition. Every record reproducible. Every licensed professional able to show — not merely assert — that they directed and reviewed AI-assisted work before it entered a sealed deliverable.
That record becomes something a carrier can ask for at renewal and a board can read years later. The methodology evolves; the commitment to audit-grade rigor does not.
The thesis
AI answers questions. This work answers the question every licensed AEC professional must answer before they sign.
Can I defend this? That's the question. Not “is the AI right,” not “is the output good” — the question that decides whether a professional engineer, architect, or licensed surveyor can put their stamp on AI-assisted work and stand behind it five years later when the E&O carrier or the license board comes asking.
AECO Shield is the record between AI and the stamp. It captures who held responsible charge of a sealed deliverable, which AI touched it, and that the professional directed and reviewed the work before it was sealed — as evidence a carrier, a board or opposing counsel can read years later.
“You can outsource your thinking, but not your understanding.”
Agents can process information and generate views of the data. But the human still has to know what is being built, why it matters, and how to direct the system. The AI is fast. The licensed professional stays accountable.
The author
Written by someone who carries the same exposure.
AECO.digital is an independent research project by a practicing professional engineer with 20+ years in the field — not a product built by an outside software team. The author seals work under the same responsible-charge rules this research documents.
Marcin Kasiak is a practicing professional engineer who researches and writes on AI in licensed practice. He holds PE, PhD (structural engineering), PMP, and IWE credentials, with 20+ years in architecture, engineering, and construction — writing about where engineering practice ends and the future begins: AI in structures, digital twins, predictive analysis, and the tools actually changing how we build.
His career has run at the intersection of structural design, project delivery, and technology adoption — the exact terrain where AI in AEC creates both leverage and exposure. He seals work under the same responsible-charge rules this research documents, which is the vantage point it is written from.
The methodology was developed over multiple years of practice observation. It is published and version-stamped, and explicit about what it establishes and what it does not: it records that a licensed professional exercised judgment, and never claims to have made that judgment for them.
This is independent research. The views expressed are the author's own analysis and the published methodology — not the positions, and not the work product, of any current or past employer or client.
The principles
Five principles the methodology runs on.
01
Records, not verdicts.
The methodology documents that a licensed professional exercised responsible charge over AI-assisted work. It does not score the tool, rate the vendor, or judge the engineering. The professional makes the call; the record shows they made it, when, and on what basis — which is the part that has to survive scrutiny years later.
02
Grounded in licensed practice.
Responsible charge, standard of care, and the record-keeping duties a state board already imposes are the anchor — not a rating scale invented for the purpose. The obligations exist whether or not AI is involved; the methodology documents how they were met when it was.
03
Version-stamped, retroactively safe.
When the methodology updates, every past record still references the version it was stamped under — frozen. Versions release on a 3–6 month cadence. The methodology evolves; past records stay valid against the version they were stamped under, because re-stamping evidence to a newer standard would rewrite what was actually asserted at the time.
04
Transparent in method.
The methodology is published in full and its reasoning is traceable: every standard it cites, every jurisdiction it tracks, and the grounds for each conclusion. Hidden methodology is not defensible methodology — a reader who disagrees should be able to see exactly where, and check the source themselves.
05
No data extraction.
The research never trains on a firm’s data. Project context stays in that firm’s tenant, encrypted at rest, with row-level security. Nothing a firm records becomes an input to anyone else’s analysis. A record built to prove accountability cannot quietly become a source of leverage over the people who kept it.