Blog
Writing for ADHD brains that build
AI-first development, neurodivergent productivity, and lessons from shipping fast.RSS
Benchmark Multi-Model A/B: Co Daje JIT Context OS, Gdy Jakość Kodu Wynosi 100%?
A/B benchmark na 5 modelach (w tym Llama 3.1 8B lokalnie na Mac mini). Prompt tokens spadły o 59-88%, latencja do 58%. Zero ściemy: jakość była identyczna.
We Gave Our Agent a Conscience and a Judge. The Plumbing Fixed More.
GAIA L1, 53 tasks: harness fixes took us from 34 to 45 correct. A conscience gate and a judge added two more, at 1.7x the time.
Same Session, Three Modes: What JIT and a Live JEV Call Actually Changed
No JIT, JIT without JEV, and JIT with a live JEV call on the same 10 fixtures. Wall clock fell to 1.57x. Tokens did not.
Your 1M-Token Context Window Can Still Forget: A Production Postmortem
How a 35k compression trigger, dormant spillover, and a 1.5k summary cap pushed an AI coding session to 437,606 tokens.
Ekspert AI dla Biznesu w Polsce: Jak Bezpiecznie Wdrażać Systemy Agentowe
Czym różni się ekspert od AI od teoretyka promptowania? Praktyczny przewodnik po wdrożeniach systemów agentowych i automatyzacji B2B.
Why We Killed Chat History in Autonomous Coding Agents (and Gained a 2.44x Speedup)
How eliminating chat history in favor of a sub-3ms JIT Epistemic Context OS cut autonomous agent task duration by 59% and eliminated error loops.
Why I Became an AI Developer After Being a Founder
I did not move from business into code to change careers. I moved because AI collapsed the distance between product judgment and execution.
From Spotted Work to Executable Automation
The next automation layer should not start with a blank workflow. It should start by observing work and discovering what deserves automation.
Models Interpret. Systems Decide.
A production AI principle I learned from finance and onboarding automation: use models for ambiguity, code for authority.
Determinism Over Prompts: How to Build AI Systems That Don't Drift
Probabilistic models shouldn't own business-critical decisions. How to build deterministic architectures for governed production AI systems.
From Logistics Founder to AI Systems Architect: Why Operations Beats Prompt Engineering
What founding Xpress Delivery taught me about building AI agents. Why real enterprise AI is an operations problem, not a prompt problem.
The Night an Agent Claimed 'Everything is Deployed': Lessons in Verification & Paranoia
Why agent self-reports are opinions, not facts. How we built strict end-to-end verification gates into our multi-agent coding workflows.
The One-Central-Brain Architecture: Why 20 Agent Tools Fragment Your Mind
Why switching between 20 trendy agent tools kills productivity. How one stateful central brain with compound context outperforms tooling fragmentation.
The One Who Is Many: How I Run a Team of AI Agents as a Solo Developer
I don't use AI. I operate it. Eight principles for running a multi-agent AI team as one person, learned by shipping real products to paying clients.
How I Built 160 Projects in 8 Months With ADHD
No CS degree, ADHD, and AI as a copilot: the honest story behind 160 projects and 4000+ commits in 8 months, including everything that went wrong.
ADHD Hyperfocus and Programming: Superpower or Trap?
ADHD hyperfocus can produce your best code or wreck your week. What hyperfocus really is, when it helps programmers, and how to steer it.
The Best Tools for Developers With ADHD (2026)
ADHD developer tools that actually work in 2026: task managers, focus aids, and coding tools tested by a dev who shipped 160 projects with ADHD.
How to Learn to Code With ADHD: A Realistic Guide
Learn to code with ADHD without abandoning it in week three. A step-by-step, AI-first approach from a self-taught dev who shipped 160 projects.
Coding With ADHD: The Complete Survival Guide
Coding with ADHD from someone living it: how to start when you can't, ride hyperfocus, survive context switching, and use AI as your copilot.