Insights & Engineering
Thinking on enterprise AI memory, governance and building systems that can prove their answers.
Decision fatigue is the real AI governance problem
Every governance tool adds another review queue. The fix isn't more human oversight. It's fewer, better decisions.
Read articleWhy symbolic AI failed for forty years, and why LLMs just fixed it
Provable knowledge always lost to the cost of building it by hand. That constraint just disappeared.
Read articleYour AI doesn't need to be smarter. It needs to remember.
Agent memory remembers conversations. Enterprise AI needs something different: a governed memory of the business itself.
Read articleAgent sprawl looks like sophistication. It's a symptom of missing memory.
A fleet of specialised agents reads as ambition. Usually it's evidence that the one problem worth solving, a shared, governed memory, never got solved at all.
Read articleWhat SAGE does with a contradiction, and what happens the next time someone asks
Most contradictions have a provable answer already sitting in their lineage. The rest get a human, and the ruling outlives the conversation that triggered it.
Read articleA silently corrected hallucination isn't fixed. It's erased.
Some automated correction tools patch a hallucination and move on, leaving no trace of what conflicted. What an auditor needs, what conflicted, what won, and why, has to survive the fix, not disappear with it.
Read articleNew York's chatbot told small businesses to break the law. Nobody was watching for the moment it went stale.
A city chatbot confidently told business owners to break laws that had been settled for years. The bug wasn't bad reasoning. It was that nothing in the system was built to notice its answers had already gone stale.
Read articleCorrecting a hallucination isn't the same as resolving it
A growing design pattern lets one AI system quietly overwrite another's wrong answer and call the flag closed. Swapping one unverified claim for another isn't resolution. It's a second guess with no record of why anyone should trust it.
Read articleAgentic AI is now exempt from model-risk guidance. The exposure isn't.
New supervisory guidance explicitly carves generative and agentic AI out of bank model-risk scope. The decisions those systems already make still need a paper trail, regulator or none.
Read articleMost teams don't trust their AI's context. The industry's fix for it stops at the easy half.
A 250-leader survey confirms the confidence gap in enterprise AI context is real. What it recommends fixes staleness, not two current sources that flatly disagree.
Read articleYour accreditation doesn't care that Brussels moved a deadline.
A stalled, untrusted AI rollout doesn't wait for a regulator to notice. The accreditation cycle that asks whether you can prove this was never on Brussels' calendar to begin with.
Read article'Governed memory' is about to mean five different things. Only one of them is ours.
Governed memory is becoming a label for four different products solving four different problems. Here's the specific, structural definition SAGE was actually built to satisfy.
Read articleGrounded in your documents is not the same as right about your business.
Retrieval fixes where an answer comes from, not whether it's still true. A corpus that disagrees with itself doesn't stop an AI from answering. It just adds a citation to the wrong one.
Read article