The work moved. So did I.
I spent most of two decades building streaming platforms. This year the last of that work moved to SaaS — and I noticed my hands had already moved somewhere else.
For most of two decades, if you asked me what I did, the honest answer was: I build streaming platforms. Locast — free broadcast TV for three million people across 31 US cities. WFMZ+ — nine live channels running on eight platforms, from web to Roku to a Samsung hanging in a Pennsylvania kitchen, at 99.9% uptime for over three years. I’m still proud of that work. It taught me what “production” actually means.
This year, the last of it moved on without me. And the strange part is: so had I, months earlier. I just hadn’t said it out loud yet.
The market restructured under the work
Between 2018 and 2024, a regional broadcaster who wanted to stream needed engineers, because the infrastructure didn’t exist off the shelf. That world is gone. Amagi, Wurl, Frequency and a dozen others turned streaming infrastructure into a subscription. Station groups consolidated. The broadcasters still standing don’t need a platform built — they need a vendor picked.
I watched this happen from the inside, and this year it reached my own door: WFMZ, the client whose platform I owned and ran, made the move to a SaaS provider. The shift I’d spent two years advising broadcasters about arrived at my own doorstep, on schedule.
There are two ways to tell that story. One is as a loss. The other is as the market confirming something my own hands had already figured out — because for the past year, almost nothing I chose to build in my own time was streaming.
What my hands were doing instead
While the streaming market was consolidating, I built two things, and only noticed the pattern afterwards.
The first is TAXCLARA — a Portuguese tax and accounting product for English-speaking residents, which I designed and shipped alone in fourteen weeks: a tax engine encoding IRS, IVA and Social Security rules; open-banking sync across multiple banks under PSD2; AI transaction categorization; a native iOS app. 705 merged pull requests, one engineer.
The second is the operational core of a US property-management company — On Q, eight thousand-plus properties across Arizona and Texas: double-entry accounting, owner and tenant ledgers, a full migration off PropertyWare onto an in-house platform, and lately, AI doing the first pass on work-order estimation.
Look at those two things side by side and the pattern is not subtle. Everything I chose to build moves money, and has to be right.
The care was never about streaming. It was about whether the thing is correct. I’ve just moved that care closer to the money.
Correctness-critical software is a different discipline
A streaming bug is a bad night. A ledger bug is a different kind of problem — it compounds quietly, survives restarts, and someone eventually makes a real decision on top of the wrong number. Building software where a wrong number is the whole problem changes how you work: invariants you make unrepresentable instead of merely catching; reconciliation as a feature, not an afterthought; test suites that exist specifically to protect the money paths; import pipelines designed around the assumption that banks will send you the same transaction twice and slightly differently.
It turns out the instinct that keeps nine channels at 99.9% for three years is the same instinct this work needs. The difference is that in accounting software, being right isn’t a service-level target. It’s the product.
The part AI changed
None of this would be a story if it took a team and two years. What makes it one is that a single senior engineer, using modern AI as actual engineering material — in the editor, in production, in background jobs that open pull requests overnight — now ships what used to need three people. Fourteen weeks from empty repository to a product with bank integrations and an iOS app is not a velocity I could have reached five years ago, and I’ve been doing this for twenty.
The judgment is still the scarce part. The AI writes a lot of the code; deciding what must be true, and refusing to ship until it is, remains stubbornly human.
What CYBIND is now
The portfolio didn’t go anywhere — Locast, WFMZ+, Ken’s Foods, all still on the site, all still true. But the practice’s edge today is software where correctness is the product: accounting, banking and tax systems, and the property platforms that run on them.
And I’ve resized how an engagement starts. Not a six-month retainer on faith — a one-week architecture audit, fixed price, in writing: your stack, the buy-vs-build calls, where AI genuinely helps, and a roadmap. Small enough to say yes to; useful enough to stand on its own.
The work moved. It took me a while to admit I’d moved with it. The website caught up this week.