I integrated github.com/safishamsi/gra… into the agent system this week. It parses a codebase with tree-sitter and produces a full dependency graph: every import, call, and module boundary, mapped deterministically, with no LLM involved.
It gives the agents grounding. When they take on a new repo, they start from the code's real structure (what actually depends on what) rather than a model's guess at how it fits together.
It's a foundation, not a final answer. The graph proposes how to divide the work across teams, but the agents can still evaluate it and decide the boundaries belong elsewhere. The code supplies the facts; the agents supply the judgment.
All in all, this cuts token usage not only on import but also while working by reducing the reliance on grep for codebase exploration.
2/
so I built a 3-tier recovery system.
1. mechanical reconcilers run every 5s, fix known wedges deterministically, no LLM.
2. a "doctor" agent that diagnoses the novel ones with judgment.
3. when even the doctor can't call it, it escalates to a human.
the rule underneath it all: a surgeon can't operate on their own brain. the doctor is always external to what it's inspecting, and the hypervisor has zero authority to unwedge itself, only to raise its hand.
1/
spent the last 48h teaching PRSM to heal itself.
PRSM runs as ONE long-lived session that never stops. great for continuity, brutal because over days it accumulates "scar damage". Agents stuck mid-task, state machines wedged in dead ends. restarting by hand just makes more scar.
12/ The long-term vision is not just a bigger model
It is an adaptive cognitive architecture: a system with perception, memory, curiosity, action, prediction, and self-directed learning built into the structure itself
A model that keeps learning because learning is part of how it operates
11/ From there, we can expand outward step by step
After bullet hell games, we can move into top-down action games, platformers, puzzle games, survival/crafting environments, and eventually more open-ended 3D worlds
Each step adds a new challenge — navigation, physics, causality, planning, memory, or long-term goals — while testing whether the same core architecture can keep adapting
1/ In order to test PRSM, we’re working on a brain-inspired AI architecture built around continuous learning, not static training
The goal is a system that can perceive, predict, remember, reason, act, and improve over time across different environments
Opus 4.8 just released with a new feature that allows us to modify the system prompt without losing token cache or restarting the session so we don't have to decide between instruction coherence and cost
This is huge for PRSM where state transitions involve dynamically swapping system prompts to communicate the new workflow
They literally want to see us win
The MCP tool dispatcher got rewritten. The prompt cache now stays warm across state transitions instead of busting on every tool surface change. Massively cheaper, massively faster.
And the skill-selection system pulls from a library of nearly 200 imported skills, picking per-agent recommendations from project facts. Every spawned agent comes per-equipped for its actual job.
Until now PRSM only ran structured workflows with explicit governance gates. We added a freeform lane. Less-governed agents for open exploration, research, consultation. They can then be promoted into a structured initiative if the work pans out.
You don't always know up-front whether a question deserves a full workstream. Freeform lets you start loose and tighten only when there's something worth executing on.
PRSM grew an apps framework. External processes (personal assistant, digest, mail bridge) now boot independently and self-register with the engine, announcing their tools. Engine stays narrow, orchestration only. Apps own their own concerns.
Adding a new capability went from "fork the engine" to "ship a separate binary that registers on startup." That's the unlock for an ecosystem.
I've been spending a lot time trying to keep up to date with the latest news in AI. Then I thought what if I automated the search and digestion process based entirely off of my interests as a twice-daily digest.
I built the entire thing with #PRSM
Subscribe if you're
3K Followers 6K FollowingWeb3 Investor | OCM OG 🦍 | NFT Strategist | On-chain ≠ just tech — it’s a movement. | Building in the open. #OnChainMonkey #CryptoNative
1K Followers 1K FollowingYour edge in crypto. 📡 Breaking news, on-chain analysis, and market insights before they trend. No shills. No noise. Just catalysts.
77K Followers 36K FollowingCrypto/Altcoin specialist🚀 | | Grow your project with me . 24 hours ⏰ Active. Marketer Admin 👨💼 | DM Me Any Business inquiries 📥
640K Followers 14K Following🌍 Passionate about the future of finance
Promoting innovative #Bitcoin & #Blockchain projects
📈 Educator in cryptoassets and decentralized finance
77K Followers 51K FollowingPumpfun/Altcoin Projects Researcher 🔍 | Boost your project with me | 🕰️ 24Hr Active Marketer Admin 👨💼 | DM Me For Promotion Services📩
#BTC #BNB #SOL #ETH
39K Followers 268 FollowingKernelBench-Hard |
i made the 12-hour CUDA course on FreeCodeCamp |
"the timelapse guy" |
books i wrote: https://t.co/2hzezpoDwQ |
@shipfr8 alum
2K Followers 1K FollowingWe're on a mission to connect, educate, and celebrate the startup community!
Grab your FREE ticket👇to Startup Boston Week 2023 right now!
663K Followers 3 FollowingLaunch a coin that is instantly tradeable in one click for free: https://t.co/5I8qHw6eDl Support: https://t.co/AodX1mIezE Download the mobile app 👇🏻
3.9M Followers 3K FollowingThe high performance network powering internet capital markets, payments, AI agents, and crypto apps. Now dealing the @wsop.