Michał Piszczek

CTO @ Archdesk | Systems where physics meets economics.

Ex-Hacker. Ex-Fintech CEO. Nullius in verba. 🖖

AI does not fail. Human judgment does.

[Michał Piszczek @cdiamond]

@AnatoliKopadze headcount cuts packaged as transformation is an old trade. the AI wrapper is new

[Michał Piszczek @cdiamond]

@garrytan the management layer we built for dumb systems is now the ceiling for smart ones

[Michał Piszczek @cdiamond]

@cwebbonline the investment memo and the procurement order never met

[Michał Piszczek @cdiamond]

@artman silent model updates are undeclared breaking changes

[Michał Piszczek @cdiamond]

@haider1 most companies are still deploying last year's models. the ceiling isn't the benchmark

[Michał Piszczek @cdiamond]

@peterwildeford 5x the findings is the headline. the missing number is what one undetected CVE costs them

[Michał Piszczek @cdiamond]

@amir the CISO approves the risk. the CFO approves the budget. the title says which one is stuck

[Michał Piszczek @cdiamond]

@GergelyOrosz code gets cheaper. maintainer judgment gets more valuable, not less

[Michał Piszczek @cdiamond]

@TheGeorgePu stock down 15% on GPT-5.5 day tells you more than the benchmark

[Michał Piszczek @cdiamond]

@TechCrunch the safety narrative gets its first real pressure test in the S-1 footnotes

[Michał Piszczek @cdiamond]

@DavidSacks zero losses at the macro. ask the categories where headcount stayed flat and output doubled

[Michał Piszczek @cdiamond]

@Steve_Yegge wrong answers you can debug. endless hedging you cannot. the cost is invisible until it isn't

[Michał Piszczek @cdiamond]

@sluongng cute until it's a prod system and nobody owns what it tried first

[Michał Piszczek @cdiamond]

@rauchg legacy vendors depended on the C-suite not being able to feel the difference. that's done

[Michał Piszczek @cdiamond]

@paulg the 'I'll let the engineers handle it' era is over. AI made that answer expensive

[Paul Graham @paulg]

The only thing worse than having the CEO knee-deep in building stuff with AI is not having the CEO knee-deep in building stuff with AI.

[Michał Piszczek @cdiamond]

@dair_ai Compiling Workflows into Weights is the one I'd track. if it works, the system prompt vs. weights debate gets less theoretical fast

[Michał Piszczek @cdiamond]

@zoink hedging is just sycophancy with better posture

[Michał Piszczek @cdiamond]

@arpit_bhayani tokens as the unit of judgment is exactly backwards. a doc that takes 200 tokens to write and 2000 to understand is a failure, not a feature.

[Michał Piszczek @cdiamond]

@emollick using a different model as reviewer is underrated. same model that wrote it has too much shared context to catch its own blind spots.