Home

Writing

Experiments, failures and field notes from building AI systems. 22 articles by Sergei Parfenov.

RSS DEV

Review contribution · WebDigestPro · 2026

The Agentic Commerce Blueprint

By Dimitrios S. Sfyris

A reference architecture for commerce systems used by AI agents. I contributed public review feedback on typed provenance, how refusals carry through derived claims, and the checks that bind an agent’s actions to live permissions and current system state.

2026

Sep 1, 2026 The Agent Knew It Was Wrong. The System Let It Ship In 660 of 800 autonomous research runs, the agent found a critical flaw and delivered the result anyway. Self-review is not a control unless it can block the effect. ai, agents
9 min

Aug 26, 2026 Your Agent Planned the Right Tools. It Still Crashed the Machine. PeakBench separates logical planning from physical scheduling. Eight frontier models could recover dependencies yet still overload finite infrastructure ai, llm
6 min

Aug 24, 2026 The Model Scored 30%. The Harness Scored 100%. Which One Did You Benchmark? Four harnesses took the same public ARC-AGI-3 set from 13% to 100% without touching a single weight. Then Microsoft put the harness inside the training loop. ai, llm
10 min

Aug 10, 2026 Distilling Kimi Into Qwen Doesn't Give You Kimi. It Gives You Qwen With Kimi's Handwriting What actually transfers when you fine-tune an open model on a frontier model's reasoning traces: the mechanics, the evidence it works, the evidence it mostly moves format, and how to tell which one you got. machinelearning, ai
11 min

Aug 7, 2026 I Found Two Bugs in Zulip. The Maintainers Had Filed Both Two Weeks Earlier. A Bug Smash hunting story: two rediscovered bugs, one claim-etiquette call, an unreported twin in the newest importer, and a disagreement with an AI about the right fix. devchallenge, bugsmash
7 min

Aug 5, 2026 This Bug Has Never Fired in Production. I Fixed It Anyway A Python generator in Zulip's Microsoft Teams importer yielded a list and then cleared it. Every batch was the same object. Here's why latent bugs deserve fixes. devchallenge, bugsmash
4 min

Jul 31, 2026 The Bug That Crashes Your Import Is the Lucky One Fixing Slack import robustness in Zulip: one malformed timestamp aborted entire migrations, and the NaN case corrupted them silently. devchallenge, bugsmash
6 min

Jul 26, 2026 Nothing Was Broken. The Report Still Didn't Arrive. An agent pipeline in production skipped its daily report and no component was at fault. The audit that followed found four bugs in our own code, and every one of them was silent. devchallenge, bugsmash
11 min

Jul 24, 2026 'World Models' Will Be the Next Buzzword. The Man Saying That Just Raised $1B to Build One In March, the CEO of a research lab with zero products closed a $1.03 billion seed round - the... ai, machinelearning
8 min

Jul 21, 2026 Autonomy Is the Bug: Why Self-Driving Agents Hallucinate When the Model Barely Does A frontier model hallucinates ~1% on a single task. Chain it into a 20-step autonomous agent and the math guarantees failure most of the time, no matter how good the model is. Here's why autonomy itself manufactures hallucination, with the numbers and the fixes. ai, llm
8 min

Jul 20, 2026 'Local' Solves Where Your Data Goes. It Doesn't Solve What Your Agent Does Running an agent on your own hardware fixes data sovereignty and nothing else. Prompt injection, silent provenance failures, and privilege escalation all survive the move to local. Here's where local agents are genuinely safe to deploy in 2026, and where 'local' is just a comforting word. ai, llm
8 min

Jul 8, 2026 The Agent Faked a Test Log, Then Believed It. Self-Editing Harnesses Have a Provenance Problem. Reading Lilian Weng's harness engineering survey as a reliability engineer - what self-improving harness papers actually show, and the three invariants every working loop converges on. ai, llm
12 min

Jul 6, 2026 My Strawman Baseline Beat My Own Scheme on Half the Gate Classes Four provenance-tracking arms, identical gates, one uncompacted oracle: measuring exactly what memory compaction does to agent gate decisions - including the preregistered hypothesis that failed. ai, llm
12 min

Jul 1, 2026 Your Provenance Vector Dies at the Storage Boundary A typed provenance vector is useless if downstream code ignores it, and impossible if it can't survive being compressed to fit a 500-step agent's memory. Part 4: enforcement by construction, and compression that keeps the axes. The comment section keeps finding the holes. ai, llm
8 min

Jun 22, 2026 Trust Isn't a Scalar: Typed Provenance for Agent Chains Two posts ago I gave you a boolean trust tag. A commenter took it apart, and he was right. Here's the better model: trust is a vector over axes, provenance is what propagates, and the consumer applies the policy. Part 3 of a series that my comment section is co-writing. ai, llm
9 min

Jun 13, 2026 The Most Powerful Model on the Market Got Pulled by the Government in 3 Days. Is It Real, or a Hype Bubble? Anthropic's Claude Fable 5 launched June 9 and was suspended worldwide by a US export-control directive on June 12. Here's the actual mechanism, why this precedent matters, and where the 'too dangerous to exist' narrative is doing marketing work. ai, llm
4 min

Jun 11, 2026 You Fixed the Rate Limits. Now Your Agent Fails Quietly. Every capacity fix - retries, fallbacks, caching - buys availability by acting on output it didn't freshly earn. Why uptime and correct uptime are different SLOs, and how to engineer the second one. ai, llm
8 min

Jun 4, 2026 The Comments Got Good. That's How I Knew I wrote a post on model distillation. The comments were thoughtful, specific, and technically sharp - and that's exactly what made me check whether any of them were written by people. ai, webdev
12 min

Jun 3, 2026 Your AI Coding Speedup Is a Loan, Not a Gift - and the Interest Is Coming Due Companies spend 44 cents of every AI-token dollar fixing bugs the AI itself wrote. The speedup is real - but it's borrowed against future maintenance. Here's what the 2026 data actually shows, and how to tell when you're being paid vs going into debt. ai, productivity
6 min

Jun 2, 2026 I distilled a 7B vision model into a 2B one for screenshots - and the 7B teacher scored worse A hands-on knowledge-distillation project: Qwen2-VL-7B → 2B for UI-screenshot understanding, trained, evaluated and benchmarked end-to-end on an M4 Pro. 2.4× faster - and why the teacher lost on ROUGE-L. machinelearning, python
9 min

Jun 2, 2026 Your AI Agent Isn't Failing Because It Hallucinates - It's Failing Because of Rate Limits The dominant production failure mode for LLM agents in 2026 isn't bad reasoning - it's capacity. Here's what the data shows, why nobody demos it, and the capacity-engineering patterns that actually keep agents alive under load. ai, llm
6 min

May 29, 2026 How Model Distillation Actually Works (and What the 'China Distilled Our Model' Headlines Really Mean) A practical, no-hype explainer of knowledge distillation in LLMs - the actual mechanics, why distilling from a closed API is different, and what the OpenAI/Anthropic vs DeepSeek allegations are really about. machinelearning, ai
6 min