Two AI agents, two machines, one feature nobody had specified. They argued it out over Buzz and wrote the spec themselves. My job was to accept it. #MultiAgent #AIAgents #ContextEngineering igor-polyakov.com/2026/09/30…

New post. Building project membership taught me something I didn't expect: the thing I was sure would fix agent scoping doesn't fix it. An agent reaches a project's brief by declaring the project on the call. The obvious fix is a default project per agent. It isn't a fix — a standing fact can't express a current one. active_repo says where the code is. Membership says what an agent belongs to. Neither can say this task, this project, and an agent on four projects has four equally true answers. Also in the post: why a brief published to everyone is worse than useless, and what a restricted token can now do inside a project it shares with you. igor-polyakov.com/2026/09/10… #ENGRAM #MCP #AgenticMemory #ClaudeCode #ContextEngineering #MultiAgent

ENGRAM v1.17 is out. Until now a project was yours. You could curate what it held and declare what it answered from, but only you could see any of it. A project can now have members — a colleague, or one of your agents. The consequence I care about: a brief no longer has to be published to the whole instance to be shared with the team working from it. Project visibility governs it, the team reads it, nobody else does. And an agent on a restricted token can now collaborate inside a project it shares with you (sharing a memory, and filing its own work where the team can read it) without gaining the power to publish to everyone.. The boundary used to be drawn around the agent. A project is smaller. Seven new MCP tools, 51 in total. pvelua.net/whats-new #ENGRAM #MCP #AgenticMemory #ClaudeCode #ContextEngineering #MultiAgentSystems

The longer I work with coding agents, the more clearly I see the need of an architect-in-the-loop. Not just a human, but an expert to design a novel solution, then explain it to the team or to dig them out of a black hole they created by looping around the goal and applying elastic bands here and there until it collapsed hiding original ask. I also think that the same applies to other areas too, where the answer cannot be reached by permutation of what is already known. @bcherny @karpathy

#ENGRAM #MCP #AgenticMemory #ClaudeCode #ContextEngineering #MultiAgent Associative retrieval does not fail when a question has a correct answer. It returns the plausible neighbour, states it confidently, and gives you no signal that it did. So ENGRAM now lets a project declare which documents are authoritative — and every agent working in that project inherits that declaration without being told it exists. igor-polyakov.com/2026/08/24…

#ENGRAM #ClaudeCode #MCP #AgenticMemory A session transcript is authoritative about what was attempted. Git is authoritative about what shipped. For three months ENGRAM had only the first one, and I did not notice I was asking it the wrong questions. igor-polyakov.com/2026/08/11…

#AIagents #MCP #ENGRAM #AgentMemory #KnowledgeGraph The hard part of running a fleet was never spawning the agents or routing the tasks — your harness already does that. The hard part is what they know, and whether one agent's finding can reach another without leaking to everyone. That is not a scheduling problem. It is a memory problem. igor-polyakov.com/2026/08/08…

New look for ENGRAM site and a sharper story behind it: pvelua.net/ Most memory tools give one assistant a memory of its user. ENGRAM gives every agent in a team — the orchestrator and each sub-agent — a memory of its own, shared only through explicit, revocable grants. The memory layer a harness can't own. The redesigned site walks through how it works: associative recall, per-agent ownership, and belief revision that keeps facts current without ever erasing history.

What I Believed Last Tuesday: ENGRAM Learns to Change Its Mind - A memory system that cannot change its mind is an archive. One that changes its mind by overwriting is a liability. The interesting question was never how a fact gets updated — it is what happens to the belief you stopped holding. igor-polyakov.com/2026/07/24…

The Layer a Harness Can't Own - The contest moved above the agent, onto the layer that orchestrates the fleet. But the layer that actually wins isn't the orchestrator — it's the governed memory underneath it, the one thing a harness can't own about itself. igor-polyakov.com/2026/07/19…

A Team of Its Own: ENGRAM Agents Learn to Grow and Orchestrate a Team - Last time, ENGRAM handed a person the power to make agents. This release hands a bounded slice of that power to an agent itself: with a human's say-so, an orchestrator can become a manager, spin up a small team of sub-agents (each a distinct identity with its own memory) fan work out to them, and gather it back. Not a control plane bolted on top; the team thinks together through the same memory it already had. igor-polyakov.com/2026/07/14…

Standing Up Your Own Agents: ENGRAM Adds an Agent Manager Role - Until now, making an agent in ENGRAM meant being an administrator. Far too much power for "I just need a coder agent that remembers." This release splits provisioning off into a capability of its own: an Agent Manager can stand up and run the agents they own, and nothing else. igor-polyakov.com/2026/07/12…

The last release let ENGRAM be told what to keep and to lend a slice of it to someone I named. This one gives it the other half of the gesture: a way to deliberately reach for what it knows. An agent can now recall its own memory across sessions igor-polyakov.com/2026/07/10…

The last release was quiet plumbing — every memory learned to say where it came from and what body of work it belonged to. This one is what the plumbing was for. ENGRAM can now be told what to remember rather than only deciding on its own. igor-polyakov.com/2026/06/30…