Insights & Articles

April 27, 2026 | Agent Grounding

Your Team’s Slack Is the New Ground Truth

Key Takeaways Chat is where most team decisions are made and where most team context lives. Ignoring it is ignoring most of the team’s institutional memory. A scheduled job at end of day reads each conversation and produces a structured summary: decisions, blockers, progress, open questions, follow-ups. Each summary chunk carries a citation back to […]

April 24, 2026 | Agent Grounding

Decisions You Do Not Have to Relitigate: ADRs in the Agent Loop

Key Takeaways An ADR captures context, decision, alternatives considered, and consequences. The full reasoning, not just the outcome. ADRs are sequentially numbered, immutable once written, and superseded (not edited) when later decisions overrule them. The brain exposes a `record_adr` MCP tool that lets any agent session create a new ADR inline as the decision is […]

April 22, 2026 | Agent Grounding

L1 Brain, L2 Atlassian, L3 Web: A Cache Hierarchy for Team Knowledge

Key Takeaways L1: a team-context brain with project-scoped retrieval, sub-50ms queries, compiled and curated content. L2: existing Atlassian estate (Jira and Confluence) reachable through the official MCP server. Slower, broader. L3: web search. Last resort, only for genuinely external information. The cache hierarchy is binding. Every agent session reads it as a non-negotiable rule at […]

April 17, 2026 | Agent Grounding

Triple-Validated Retrieval: How We Stop the LLM From Making Things Up

Key Takeaways Single-layer hallucination guards miss things that the layers above and below would catch. Defence in depth is essential. HaluGate runs at compile time on wiki articles, extracts factual claims, cross-references each against the source material, and flags risk levels. Citation verification runs at serve time on every synthesised answer, dispatches each claim to […]

April 15, 2026 | Agent Grounding

Yesterday’s Gospel: Why Retrieval Should Age Out

Key Takeaways Pure semantic relevance does not encode time. A retrieval system without decay will surface old “we always” content alongside last week’s revision, equally weighted. Half-life decay is a multiplier on the relevance score based on the chunk’s last-modified timestamp. A 90-day half-life means content from 90 days ago has half the weight of […]

April 13, 2026 | Agent Grounding

Save-On-End Is Broken: Why AI Sessions Should Persist Turn by Turn

Key Takeaways The clean-shutdown fallacy: save-on-end persistence breaks the moment a real-world event (crash, force-quit, network drop) bypasses the shutdown hook. The trickle pattern writes the assistant’s most recent turn to disk before the next turn fires. Worst-case loss: one in-flight turn. A periodic re-summariser turns raw turn-by-turn transcripts into topical chunks (decisions, blockers, progress, […]

April 10, 2026 | Agent Grounding

The Repo Matrix: A Living Document That Stops AI Agents Hallucinating Your Codebase

Key Takeaways The repo matrix is a structured snapshot of the codebase: endpoints, schema, dependencies, env vars, framework. Mechanically scanned, not LLM-generated. Cheap, fast, deterministic, reproducible. The same scan over the same tree gives the same answer every time. AI agents query the matrix before suggesting code. New endpoint suggestions get checked against what already […]

April 8, 2026 | Agent Grounding

Why Your Team’s AI Memory Should Ship as MCP Tools, Not Just an API

Key Takeaways MCP exposes server capabilities as discoverable tools that any compatible AI agent can call without custom integration code. The five tools every team-context memory should expose: `memory_search`, `memory_ask`, `memory_expand_ref`, `memory_session_list`, `record_adr`. Adoption cost on the agent side: zero. The agent runtime already knows how to discover and call MCP tools. No SDK, no […]