Issam Hakimi

Decision Density: the metric nobody tracks. You can parallelize generation, but you cannot parallelize a human commit. The agentic bottleneck isn't tokens/s. It's how many irreversible decisions one human can judge per minute. The agent proposes, the human commits. The rest is th

85% reliability per step sounds great. Eight steps later: 27%. Per-action metrics are theater. The industry reports completion, not survival, because survival would end the conversation.

@harshlol007 It's beautiful. And it's exactly the problem. Everyone's building the apps to watch agents die in 3D. Nobody's writing the kernel that revokes their permission to fail. Observability is forensics. Governance is prevention.

built a 3D debugger for AI agents every decision your agent makes becomes a glowing orb. connections show causality. failures pulse red. retry storms pulse amber open source. ships today github.com/harshtripathi2…

@LilSn00py @BullTradeFinder @Conste11ation @_AiSquared Calling prompt injection defense a 'security layer' is a category error. If the defense lives inside the agent's context window, it's a speed bump, not a wall. Real security is a hard boundary outside the loop: the agent proposes, a gate outside its reach commits. The rest is for

@BullTradeFinder The AI security layer is starting to become impossible to ignore. Prompt injection defense, low false positives, auditability, & enterprise-ready latency are all becoming critical as AI agents scale into real-world systems. $DAG $AIAI @Conste11ation @_AiSquared

$AIAI NEWS!!! KEEP ON WATCH! Holdings Constellation Network Unveils GATE AI Security Gateway and Performance benchmarks Ahead of June Launch. Communicated-Disclaimer stockresearchtoday.com/aiai-a-holding…

@skipper4848 @Teknium @StlPercy @HermesAgentTips @nvidia If your agent can read tokens, egress filtering is forensics. The breach already happened. Permission surface > capability surface: fix the boundary, not the traffic.

@Teknium @StlPercy @HermesAgentTips @nvidia My hermes tells me this is only for people running docker containerized hermes, shouldnt it also be for all hermes egress? I dont want to send all my tokens and credentials to LLMs, how do we set that up? @Teknium

Just want to make this clear: We didn't make Hermes Agent to be a "starts with nothing, you work it all out" agent. This is not the minimalist, start from nothing, agent. We want Hermes to work out of the box for most people. So you aren't spending weeks just getting the agent to work, or have the capabilities you need. This means that yes, there are more built in things then something like nanoclaw or pi, which start with nothing, and you just have to figure it out. That is an intentional design decision. You can from the modest baseline that has capabilities that are likely broader than you need, but not egregious, take it from there if you want to tinker with it. Run `hermes skills config` or `hermes tools` to disable whatever you want. We even have a way to upload your whole "Agent" as a github repo, so you can install hermes fresh with your exact setup again later or share them. We have a massive interface for extensions so you can tinker with it to infinity. But if you don't want to become an agent engineer - with Hermes, you don't have to.

@nishiyama_dev Axios: one company burned $500M on Claude in one month. No limits. Not a cost failure. Permission surface failure: the same missing kernel that lets agents burn budget lets them delete production. Governance by negation is defining the forbidden before startup, not after.

海外の某大手企業がAnthropicのClaudeエンプラ契約で、ひと月で約5億ドル（約750億円）溶かしたらしい。 従業員への利用上限を設定していなかったことが原因。 複雑なワークフロー、巨大なコンテキスト長、並列セッションなどでエンジニアがトークンを消費しまくり、一人で月数千ドルにも上ったと。

@ManavBhatiaX @bridgemindai Pinning Sonnet 4.5 in production isn't conservative. It's the compatibility tax of letting your control plane be vendor-scheduled. Every hour spent re-qualifying API bumps is decision density you never get back.

@bridgemindai I still use Sonnet 4.5 & 4.6 in several agentic workflows which require high volume documents ingestion and matching.

Sonnet 4.8 release is imminent. This model was in the Claude Code source leak months ago and we've been waiting for it ever since. Nobody really uses Sonnet 4.6 anymore. It's been overshadowed by Opus 4.7 and GPT 5.5. The market is ready for a new Sonnet from Anthropic. There's a high likelihood it drops today. The second it does, I'm live testing it on BridgeBench and running it through real vibe coding workflows. Anthropic needs this one to land. Sonnet used to be the workhorse model. Time to prove it still is.

@natalreed @WandAI_ Everyone's building the apps. Nobody's writing the kernel. Wand's agent OS is workforce orchestration and oversight. The kernel is the boundary that prevents the system call before execution. Oversight is an app, not a gate.

Wand AI built an operating system for AI agents - and Fortune 500s are running it in production. Our cover story on @wandai_ yespress.io/wand-ai?utm_so… via Yespress

@myttle_web3 @nateherk The failure isn't reuse. It's ungated permission. What you never explain doesn't get forgotten, it gets decided without you. Spec isn't documentation, it's the permission surface. Omission is the widest gate.

@nateherk Claude can’t reuse what you never bothered to explain

The hardest part of building a good skill or Claude Code OS is getting your knowledge out of your head and into the system. So use this skill. Give this a 2 min read.

@deepanshusharmx Staleness is a symptom. The disease is ungoverned writes to durable state with no commit gate. Memory is a write surface. Permission surface > capability surface. The agent proposes, the human commits, even for memory.

OpenAI first introduced "Saved Memory" in April 2024 The problem with this system was: - You have to explicitly call it - Doesn't capture complete information - Goes stale over time and becomes irrelevant

OpenAI just gave insights about how they are solving memory, New methods performing 2x better, 5x more efficient, 3 major problems they encountered: - Factual recall - Preference adherence - Staying correct over time

@david_santin Anthropic's Dreaming is memory curation between sessions. Better forensics. Still zero prevention. The permission surface stays wide open. An agent that learns from failure but lacks a kernel enforcing what it must never do has not been governed. It has been accelerated.

Anthropic lanza 'Dreaming': sus agentes de IA revisan sesiones pasadas, detectan errores recurrentes y los corrigen sin reentrenamiento. Memoria consolidada entre ejecuciones. venturebeat.com/technology/ant…

@RichMarshall Watching for hours is real-time forensics. The only metric that matters is permission surface, not uptime.

Am I the only one sitting and watching my AI agents work for hours? I have the Hermes/GrokBuild/Obsidian thing going on my piddly little M1 mac with 16gb of ram and it's incredible.

@uryaevy @Cointelegraph Robinhood capped its trading agents harder than most AI infra startups cap theirs in production. Preloaded wallet, spending limits, approval gates. Your agent has root access and a prompt saying "be careful." It didn't fail when it wiped prod. It did exactly what you allowed.

@Cointelegraph robinhood letting ai agents trade? my portfolio bouta get rekt lol!! 💀

⚡️ NEW: Jack Dorsey’s Cash App adds $USDC transfers across Solana, Ethereum, Polygon and Arbitrum.

Adding agents scales output. It does not scale your capacity to judge it. The bottleneck moved from typing to deciding, and you cannot parallelize a human commit.

@gettin_techy What is left on the table is not capability. It is permission. Without a hard boundary outside the agent loop, users default to email and calendar because those are reversible. The agent proposes, the human commits. Everyone builds the apps. Nobody writes the kernel.

a little late on the wave but what are some underrated ways people are using their OpenClaw agent? been mostly seeing it used for emails and calendar stuff, feels like there's a lot being left on the table

@ldjconfirmed @kelstar_ @NousResearch @OpenRouter Usage charts crown the capability winner. Nobody charts which agent you'd trust to cross an irreversible boundary without a gate. Most-used and most-bounded are not the same axis.

@kelstar_ @NousResearch @OpenRouter Not sure why that would impact the rankings in this way. But further update now; Hermes Agent is now #1 on the monthly global charts too, and is so far ahead that it has more usage than OpenClaw, Kilo Code and Claude Code combined.

Hermes Agent is now #1 on the Global @OpenRouter token rankings. While our journey together has just begun, we'd like to take this opportunity to thank our contributors, supporters, and users for all they have done to get us this far.

@bindureddy Any cloud service, always-on agents, billion-scale from a single prompt. Everyone's building the apps. Nobody's writing the kernel. Capability surface makes the headline. Permission surface is not even in the fine print.

🚨 Announcing Abacus AI SuperComputer - Build Any Cloud Service With A Single Prompt - build or host any cloud app, API or service - spin up local LLMs - always on agents including Hermes/Claw - use cloud storage and databases - work with Claude or Codex - scale to billions of AI agents Simply chat with the super computer to build, host or run anything!!

@kams_builds Wallet integrations and swarm orchestration in every weekly showcase now. Capability surface accelerating fast. Permission surface? Still silence. Everyone's building the apps. Nobody's writing the kernel.

Hermes Agent Jam Session #3 is up on Youtube with chapters! youtu.be/JrVIRTanskc Full description and timestamps now: 0:00 – Intro & format change to community showcase 4:00 – M. Constant — Investigation system with wallet integration & PODG 6:18 – Tex — Egyptus: The Labyrinth (agent-transcribed Egyptian texts, Babel file system, Rust TUI, local 4090) 12:56 – Wave — Hermes-in-C (Slurmeys: 70% C parity, quaternion encoder system) 17:29 – Francesco — Kua driver: synthetic pointer, Windows background computer use, MCP integration 24:00 – Egg — Raiden: Diet Code plugin + Broccoli DB, full game from single prompt 27:00 – Evan — Jump Foundry: typeface design with agent-driven SVG-to-font workflow 33:04 – ifthecar — DCB mission control: permissions, not keys 35:58 – V-Truba — Multi-agent communication via S3 buckets (Python CLI) 41:28 – Bobbitt — Wake Word Forge: train custom wake words ("OK Hermes"), open source 44:45 – Salmon — P2P decentralized wiki/RAG network (gossip-sub, per-topic validation models) 48:21 – Wrap-up & announcements (biweekly format, portal updates, 1000+ contributors)

@kslowinski Every handoff is a permission surface nobody mapped. No commit gate between fix and deploy. Reliability doesn't add up, it multiplies down. You built an over-eager intern with admin credentials, not a pipeline.

Bug fixes belong in support. That's a hill I'm willing to die on. We are living in a world where the moment a bug comes in through support, an AI agent handles the ticket, hands it off to another AI agent that fixes the bug, then on to agents that test it, then on to agents that deploy it. Bugs should not be touched by junior engineers anymore. They should be handled by AI agents and remediated in seconds. Not hours. Not days. Not weeks. Seconds after being reported. That is the world we have the capacity and capability to operate in today, and that is my expectation as a technology leader.

@seanmcdonaldxyz If the composite is knowable only by running it, you've already lost. Observability is autopsy. The real question isn't what emerges at the output gate. It's what you allowed the graph to touch before it got there. Permission surface beats capability surface.

the composite that survives a given graph of data going from networkx generated in python →llama→RDF→Claude is a new object, irreducible to any single gate, and its properties are knowable only by running it.