Insights
Arguments, not thought leadership.
Everything here is something we've had to say out loud in a real meeting, usually more than once, or a system we built and can still open. No gated PDFs, no newsletter wall.
Arguments
What we end up saying out loud in a buying conversation, usually more than once.
- Engineering6 min read
What breaks when you replace an AS/400 while the plant keeps running
The hard part is not reading the RPG. It is that the machine cannot stop, the cutover window is a holiday shutdown, and four specific things fail that nobody budgets for.
- Systems7 min read
EDI 214 in practice: why your ETAs are wrong
The status feed is not a truth feed. Almost every bad ETA we have been asked to fix comes from four failures that are visible in the message itself.
- Strategy4 min read
“Work made for hire” does not mean you own it
The phrase appears in nearly every development contract and it is doing far less work than the people signing it believe.
- Delivery5 min read
The statement of work is the contract
The legal terms get the lawyers and the weeks of redlining. The document that actually decides what you receive is usually written the night before signature.
- Strategy4 min read
Build versus buy: the only test that actually settles it
Most build-or-buy debates are decided by whoever presents last. There is a better question, and it takes about ten minutes to answer.
- Engineering4 min read
Why your legacy rewrite keeps getting cancelled
It isn’t a failure of nerve. Executives are correctly refusing to accept a risk profile nobody should accept, and there’s a way to change the profile.
- Data2 min read
The three-numbers meeting, and how to end it
When finance, sales and operations bring different figures for the same month, the problem is never the dashboard tool.
- AI3 min read
AI with a scoreboard: which projects pay for themselves
Three conditions predict almost every applied-AI success we’ve shipped. Projects missing any one of them tend to fail in the same way.
- Delivery2 min read
An estimate is a range, or it’s a bid
Single-number estimates communicate false precision, and everyone involved knows it. Here’s what we give instead.
- Engineering3 min read
Offline-first is a product decision, not a technical one
Field software fails for a reason that has nothing to do with engineering skill: someone assumed the network.
Engineering write-ups
Systems we built end to end, walked through in full, with the parts that did not work left in.
- Systems6 min read
Consensus is not the hard part. Reconfiguration is.
Every distributed-systems reading list ends where the real engineering starts. What breaks in production is the day the topology changes.
- Engineering5 min read
What building a compiler teaches you about modeling anything
Pick the wrong intermediate representation and every later stage pays for it. That is not a compiler problem; it is the problem.
- Data7 min read
In applied machine learning, the model is the tiebreaker
We built the full pipeline from raw network capture to a live detector, measured where the time went, and the answer was not the modeling.
- Engineering5 min read
The shortest path is usually the wrong one
A routing problem taught us more about requirements gathering than about geometry. The algorithm was never the hard part; the objective was.
- Engineering5 min read
The count is the contract: what semaphores still teach
Nearly every synchronization pattern is the same two operations with a different starting number. Learning that changes how you read concurrent code for good.
Next step
Disagree with any of this?
Good, that’s usually the start of a useful conversation. We’d rather argue with you before an engagement than during one.