Blog
All my articles and thoughts.
-
What Survives Compaction Is the Real Context Window
• 6 min readJune's research reframes context management: the discard step is now where both agent quality and safety quietly leak.
-
The Cheapest Agent Upgrade Is a Stop Condition
• 5 min readMid-2026 data keeps pointing the same way: bounding an agent's loop beats unleashing it. Turn limits and budgets buy more than a bigger model.
-
Ten Agents, Three Merges: June's Tooling Fixed Fan-Out, Not Review
• 5 min readThis month's agent tools made spawning parallel coding agents trivial. The constraint moved to the merge decision—and that doesn't parallelize.
-
Agent Security Moved to the Action Layer
• 6 min readRuntime authorization — intercepting tool calls before they execute — is becoming the real security boundary for agents, and a standard is forming fast.
-
Computer-Use Agents Crossed Human Parity. They Still Click Too Much.
• 6 min readFrontier models now beat the human baseline on OSWorld-Verified — but the benchmark just got rebuilt, and the architecture quietly shifted off pixels.
-
Your Agent Catches Everyone's Mistakes But Its Own
• 5 min readNew research says self-correction fails because of the role label on the claim, not the claim's content. The fix is structural, and cheaper than you think.
-
Your Agent's Benchmark Score Is an Experiment, Not a Fact
• 6 min readRecent work shows a single agent leaderboard number is wrong three independent ways: it's noisy, it's overfit, and the judge measuring it is unreliable.
-
Agents Are Learning the Memory Policy You Used to Hand-Code
• 5 min readA June 2026 wave moves the store/evict/retrieve decision from heuristics to a trained policy, and pushes consolidation into an offline sleep phase.
-
Injection Stopped Being a Single-Turn Problem
• 5 min readOnce agents got long-term memory, a one-time prompt injection could survive across sessions. Mid-2026 research shows both the attack and the defense moving up the stack.
-
The Agent Stopped Waiting to Be Asked
• 6 min readJune 2026 mainstreamed always-on agents that listen to event streams instead of prompts — and that one change breaks the trigger, trust, and latency models all at once.