Pinecone proposes a persistent knowledge layer that agents can consult for how an organization actually works. Its presentation distinguishes retrieving a particular contract from recovering institutional knowledge, such as who owns a responsibility or which principles guide decisions. The latter can be distributed across many documents.
Keeping that knowledge current is central to the proposal. A leadership change may be understood immediately by employees while older documents still name the previous leader. The knowledge layer needs a way for the current fact to take precedence, so repeated retrieval does not keep reviving outdated information.
Key takeaways · AI-distilled
Liberty splits enterprise AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → knowledge into three kinds: general (absorbed in training), specific (contracts/codebases, handled by retrieval for years), and tribal (how things get done), the last of which can't be pulled from any single document.
His mental model for an enterprise agent: a brilliant new hire who knows nothing about the company and restarts from scratch every session, which a persistent, specialized knowledge layer that all agents query is meant to fix.
The hardest requirement is staying current: replace a CEO and the whole company knows by morning, but thousands of documents still name the old one, so the new fact has to override all of them, not just get added alongside them.
Rather than answering directly from retrieval, Liberty's system uses a 'runtime coding agent' that writes and runs code at query time, iterating like a notebook until the answer holds, cutting their tooling prompt from roughly 150,000 tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → to under 1,000.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
videoVoice AI Where One Wrong Question Breaks the Rules - Shiv Indap, Upstart & Brooke Hopkins, Coval
video78% Wanted This AI Feature. Almost Nobody Used It - Supriya Subudhi, Bumble
videoWhen Running Your Own AI Models Beats Paying for APIs - Steven Latré, imec
videoHow to Decide What Your AI Agent Should Remember - Wojciech Pluta, Oracle
videoIf we want them to do Knowledge Work, design them as Knowledge Agents — Benjamin Clavié, MixedbreadAI Engineer
videoYour company brain will leak secrets: how we stopped it for big banks — Tanmai Gopal, PromptQLAI Engineer
videoFrom coding to Knowledge work agents — Karan Vaidya, ComposioAI Engineer