wickedagile

The best code tells a story. The best stories have architecture.

Most people don't build both.
Nobody told them they could.

the work → the writing →
wickedagile — session
❯
INCIDENT LOG · MEDIUM/@MIKE.PARCEWSKI

yes, and...

PRIORITY · LIVE

The Model Is a Rounding Error. Engineer the Harness.

● Jun 2026 · 9 min read · SOURCE medium
OPEN DISPATCH ↗
DISPATCH LOG MEDIUM ↗
WORKSPACE · 9 PACKAGES · 3 DEPLOYED

shipped.

Five chapters · ~30 years

about.

One career, read top to bottom. The same problem getting clearer with every chapter.

Chapter 01 — Early Years

Lucky enough to be in the room early.

I've been building software since before the commercial internet really had a name for what we were doing. My first real job in tech was helping put travel booking online. We had built a white-label platform for a number of large companies — pulling real-time pricing from SABRE, the whole architecture. For Expedia, we got to do something different: I led a two-person team and we designed and built their version from the ground up. I had nothing to do with Expedia's success after that. I was just lucky enough to be in the room early.

Chapter 02 — Scale

The tolerance for failure got much smaller.

From there, the work moved into financial services, where the problems got more complex and the tolerance for failure got much smaller. Syndicated data products. Compliance-grade delivery. Global market launches. Clients like UBS and Singapore Exchange who expected things to work the first time.

Chapter 03 — Architecture

Was the architecture surviving contact with reality — or just looking impressive in a slide deck?

Over time, the work shifted from building products to shaping how large organizations build them. I have always tried to stay close enough to the code and delivery teams to know whether the architecture was surviving contact with reality or just looking impressive in a slide deck.

Chapter 04 — AI

Not a model problem. An organizational one.

The last several years at Accenture have been focused on AI, specifically the gap between what AI looks like in a demo and what it takes to make it work in production. That gap is wider than most people want to admit. Data is fragmented. Systems are older than the teams maintaining them. Workflows everyone agreed to automate are still held together with undocumented tribal knowledge. Getting AI to work in that environment is not just a model problem. It is an organizational one.

Chapter 05 — Reckoning

Agents just made it obvious.

I have been wrong about plenty of things over the years. I thought the hardest part of building software was writing the code. Then I thought it was the architecture. Then the process. Now I think the hard part is making an organization operable by something other than the organization itself. That problem was always there. Agents just made it obvious. I am still figuring it out, same as everyone else.