PinnedInLevel Up CodingbyJaroslaw Wasowski·Jun 13Stop Treating Agent Memory Like a Cache — It’s a Security LayerA production agent’s memory is an authorization layer, not a cache. Relevance is not permission — where to insert a deterministic gate…A response icon10A response icon10
PinnedInLevel Up CodingbyJaroslaw Wasowski·Jun 11Formal Verification in Spec-Driven Development — Enterprise LevelIn regulated enterprise, a specification becomes an executable contract. Why AI-generated code must be proven with formal methods, not…A response icon6A response icon6
PinnedInLevel Up CodingbyJaroslaw Wasowski·Jun 18The End of Coding Is Not the End of Developers — and That Distinction Decides 2030SDD and AI agents don’t end the developer profession — they shift its value from code to intent and verification. Here’s the competency map…A response icon14A response icon14
PinnedInLevel Up CodingbyJaroslaw Wasowski·May 11Stop Writing Specs. Start Writing Facts. The Entire SDD Movement Is Already Obsolete.One test survived Sonnet 3.5, 3.7, 4, and Opus 4.5+. The spec needed four reinterpretations.A response icon50A response icon50
InLevel Up CodingbyJaroslaw Wasowski·22h agoSpec-Driven Development — Developer Trust in AI and the Cost of Verification: 2026 Survey, METR…The 2026 survey data fits a simple model: trust grows when checking is cheap. Inside: the verifiability score formula and a mini-spec
Jaroslaw Wasowski·1d agoRewriting Code with an AI Agent Fleet: Sharp Oracle, Judge Protection, and Model AttributionAn oracle that says “no” and that agents can’t write to decides what survives. Pick it first, then judge the model
Jaroslaw Wasowski·2d agoSpec-Driven Development — Breaking Changes Start in the Spec: Requirements Semver, Blast-Radius…When an agent regenerates code from a spec, one word changes the contract. Requirements semver, a diff checklist and a changelog templateA response icon3A response icon3
InLevel Up CodingbyJaroslaw Wasowski·3d agoAI Agents in Production — AI Agent Drift: Harness Fingerprint, Trajectory Distributions, and…A step budget stops one run, while drift shifts the distribution. A harness fingerprint, a frozen holdout, and bisection name the componentA response icon2A response icon2
InLevel Up CodingbyJaroslaw Wasowski·3d agoSpec Passes, Product Fails: The Scope of Proof in SDD, Coverage, and the Evidence MatrixSpec compliance proves only what the spec says. An evidence matrix for what your tests, coverage, and contracts do not prove
InLevel Up CodingbyJaroslaw Wasowski·3d agoClaude Code Mods: Deny Rules, claude plugin validate, and a Mod Review Gate for TeamsA mod runs inside the harness process with your permissions. Here is a review gate you can run today on a private plugin marketplace
