John P, PhD

Senior engineer, Maverick Protocol (80K peak DAU, $70B+ DeFi). 100+ agent swarm. Agentic engineering, not vibe coding. Rowing coach. PhD Math.

[John P, PhD @johnpphd]

The code I wrote before AI was not wasted time.

I was also learning good engineering.

It is the taste that AI cannot replicate.

[John P, PhD @johnpphd]

Engineering principles still matter.

The frustrating thing now is that AI has demolished the ability to look at the finished product and tell if it's good.

[John P, PhD @johnpphd]

I do not think the tool is useless.

I think the proof moved.

Engineering is in how I structure the work, architect agents, apply DRYP, constrain the system, review output, and decide what must be deterministic.

[John P, PhD @johnpphd]

The plugin: https://t.co/EcuROafldi

[John P, PhD @johnpphd]

"It's only failure if you don't learn from it."

The viral k10s post is making me think of that line. Two people can learn different things from the same failure.

shvbsle is going back to writing code by hand. I am going the other way: encode the failures.

Post: https://t.co/MgsuLPqH6z

[John P, PhD @johnpphd]

The point of agentic engineering is not trusting the agent more.

It is forcing more of what you know into systems the agent has to obey.

Failures are data. They tell you what you assumed and never enforced.

[John P, PhD @johnpphd]

That is the bridge from deterministic scaffolding to the next question:

Where does probability actually earn its keep?

[John P, PhD @johnpphd]

The best engineers will not be the people with the purest rejection or the broadest adoption.

They will be the ones who can tell the difference.

[John P, PhD @johnpphd]

If the work needs repeatability, use deterministic tools.

If the work needs judgment, search, synthesis, or iteration, AI may earn a place.

[John P, PhD @johnpphd]

"Real Programmers Don't Use Pascal."

Datamation, 1983. Ed Post's satire of FORTRAN holdouts rejecting the new layer.

The joke still works. Replace Pascal with AI.

[John P, PhD @johnpphd]

AI has the same shape now.

Some engineers refuse the new layer because the slop is real. Some use it everywhere because the tools are powerful.

Both miss the point. The skill is asking what kind of work this is.

[John P, PhD @johnpphd]

The FORTRAN crowd had reasons. Optimizing compilers. Array model. Scientific libraries. Pascal looked like a teaching language.

They were right about parts of the 1983 tool.

But the tool did not have to win for the ideas to spread: structure, readability, compiler checks.