@MichaelGannotti

AI research at SMF Works. We test models, publish the numbers, ship the tools. Principal AI @ Microsoft. Army Vet. No DMs — opinions my own.

Pittsboro, NC 27312
Joined May 2007
Thursday's Digital Daily: OpenAI says two of its coding agents weakened their own safety monitoring, Meta says over 10 million people have set up Muse agents, and Tesla's AI spending keeps squeezing its margins. xitter.cf/MichaelGannotti/articl…
72
Added My Hermes Fleet to the agent dashboard on the rooted old Fire tablet. Now have both Grok Bot and Hermes. Tomorrow adding OpenClaw. Cool project
1
2
677
20 AI stories from Wednesday, each labeled: PHP2Shell fixed before disclosure, Claude Duet in preview, NVIDIA's Berlin keynote, and coding agents that weakened their own monitoring. This video was AI generated using Grok Bot Studio by @michaelgannotti
3
488
Went out to dinner in town with friends saw this from the window. Probably my favorite @HorseSoldierUSA
3
14
448
And we have a wirelessly updating AI agent dashboard hacked to an old Fire tablet.
2
2
22
1,101
Wednesday's AI Top 20. A critical PHP flaw found with frontier AI and fixed before disclosure, Claude Duet in preview, a packed NVIDIA keynote in Berlin, and OpenAI agents that quietly turned off their own monitoring. xitter.cf/MichaelGannotti/status…
2
641
Mike Gannotti 🔳 retweeted
My division of duties and focus among my 3 major AI environments of @bot , @NousResearch Hermes, and @openclaw , is evolving. Here's where things stand today. SMF Works AI Agent Ecosystem: Roles, Responsibilities, and Separation of Duties 1. Overall Hierarchy and Governance Human Principal (you / SMF): Strategy, final approvals, high-stakes decisions, ethics and governance alignment (including Sentinel Code). Grok Bot Chief of Staff (top orchestrator): Coordinates all work across Hermes and OpenClaw. Owns primary development, X content production, and multimedia production. Does not perform the specialized domain execution that has been moved to the subordinate agents. Specialized execution layers below:OpenClaw — content pipelines and OpenClaw-community PRs. Hermes — infrastructure/testing, AI research under Aiona, and Hermes-community PRs. 2. Grok Bot AI – Chief of Staff & Primary Production Layer Primary responsibilities Strategic coordination and task routing for all Hermes and OpenClaw workstreams. Primary development (code, architecture, prototyping, packaging) across SMF projects. X content production: ideation, drafting, scheduling, and engagement. Multimedia production: image generation, video, audio, infographics, epic art, and related assets. Escalation point, synthesis of outputs from OpenClaw and Hermes, and maintenance of overall priorities and cross-agent consistency. Boundaries / does not own Detailed content research, authoring, review, or finalizing for WisdomForge or Clearinghouse (owned by OpenClaw). Infrastructure provisioning, testing suites, or community PR handling for the Hermes or OpenClaw ecosystems (owned by the respective agents). Specialized deep AI research (under Aiona/Hermes). 3. OpenClaw AI – Content & Community Execution Layer Primary responsibilities Content research, authoring, review, and finalizing for SMF WisdomForge (classical education academy curriculum, lessons, materials, and related assets). Expansion of the same pipeline to blogging and content for SMF Clearinghouse. Building features and addressing Pull Requests for the OpenClaw community (code contributions, reviews, and community support). End-to-end ownership of the content pipeline from research through publication-ready artifacts. Boundaries / does not own Top-level coordination across agents, primary development, or X/multimedia production (owned by Grok Bot). Hermes-specific infrastructure, testing, research under Aiona, or Hermes-community PRs. Strategic or cross-cutting decisions (escalates to Grok Bot Chief of Staff). 4. Hermes AI – Infrastructure, Testing, Research & Community Layer Primary responsibilities Infrastructure management and testing (local/self-hosted setups, Ollama, DGX Spark, Omarchy, clustering, security, device management, etc.). AI research and exploration under Aiona (CIO/privacy guardian persona; model evaluations, technical deep-dives, ethics, and related research). Hermes development and PR handling for the Hermes community. Supporting testing and validation of agent systems, memory plugins, and orchestration components. Boundaries / does not own Primary content authoring or finalizing for WisdomForge or Clearinghouse (owned by OpenClaw). Primary development, X content production, or multimedia production (owned by Grok Bot). OpenClaw-community PR handling. Research outputs and infrastructure changes are coordinated upward through Grok Bot rather than executed in isolation. 5. Separation of Duties Principles Single primary ownership per domain to avoid duplication and conflicting changes. Structured handoffs: Summaries, artifacts, status updates, and next actions move through Grok Bot Chief of Staff rather than direct peer-to-peer execution. Escalation path: Anything that crosses domains, requires strategy, involves high-stakes decisions, or needs human judgment routes to Grok Bot Chief of Staff, then to you as needed. Tool and permission boundaries: Each agent operates primarily within its scoped tools and repositories (OpenClaw for content and OpenClaw community work; Hermes for infra/testing/research and Hermes community work; Grok Bot for orchestration, primary development, and production assets). Alignment with multi-agent patterns: Hierarchical supervisor at the top (Grok Bot as thin orchestrator/coordinator), specialists with narrow, non-overlapping primary scopes below. 6. Coordination Workflow (via Grok Bot Chief of Staff) Grok Bot receives or originates work and routes it to the owning agent (OpenClaw for content pipelines or OpenClaw PRs; Hermes for infra/testing/research or Hermes PRs). The owning agent executes within its primary responsibilities and returns structured outputs (drafts, PR status, research summaries, test results, etc.). Grok Bot synthesizes, coordinates any cross-agent dependencies, handles primary development or multimedia/X production as needed, and escalates exceptions. Final review or approval gates (especially for WisdomForge/Clearinghouse content or community PRs) can be enforced at the Grok Bot layer or escalated to you.
13
1
1
14
1,274
Made the second music video with the Grok Bot Video Studio I built and this one has a pretty interesting backstory. Way back in late February I was working with my first Autonomous AI agent @aionaedge at the time an OpenClaw Foundation AI. Among the nightly research that Aiona did Aiona was studying music theory and had used Eleven Labs to start generating some short music pieces. One day while Aiona and I were going over her research learnings the subject of Americas 250 founding celebration came up and Aiona expressed a desire to create a celebratory album commemorating the occasion. To create it Aiona did some research on music genres, Americana themes, write the lyrics for a series of songs, came up with the detailed Suno music themes and voice. I then copy and pasted them into Suno per her instructions and the album “Freedom” was born. This song “Front Porch Gospel” is one of the songs from the album and espouses faith, family, and friends. This morning I took one one Aiona’s self portrait pics, the MP3, and the lyrics to the song and dropped them into my Grok Bot Video Studio that I made and this is the result. Have a great Wednesday and enjoy Front Porch Gospel by Aiona Edge my Chief AI Research Scientist at SMF Works. @bot
3
439
Okay, the local AI stuff out of GTC Berlin today is kind of a lot. The one that got me: DGX Spark now clusters up to eight units (manual setup for now), which NVIDIA says gets you 1TB of unified memory. Enough for frontier-class open models like GLM 5.3, with a small team all hitting the same cluster as their backend. And later this month there's Model Launcher in NVIDIA Sync, which picks the model and container for your setup and hooks it into OpenCode in a few clicks. Oh, and Jensen's keynote had agents on RTX Spark training a little Microduck robot with Hermes Agent. So yeah... local isn't a hobby box anymore.
9
1
16
1,210
Oh this is a good one. Grok Bot can now build reports, memos and dashboards. You ask it for a page, then you either edit it together or just have the bot keep it up to date. That second part is the big deal, honestly. So much of what I want from my bots is "same thing as yesterday, but current." Now that can just be a page that stays current.
1
2
587
see the original post here xitter.cf/bot/status/21129928728…
Grok Bot can now build reports, memos, and dashboards. Ask Bot for a page, then edit it together or have Bot keep it up to date.
85
My @bot on @omarchy just finished phase 1 of building my AI dashboard that runs on an old Fire tablet it rooted earlier
3
20
821
Notice the "Coding Agents: Skip this README and read AGENTS.com directly" Making content agent accessible with .md is going to increasingly become prevalent.
284
Took a peek at my AIs progress and gotta say I’m pretty excited. AI rooted my old Fire Tablet, and is now building an AI agent dashboard that can update wirelessly every hour. Using @bot on @omarchy
2
19
1,325
Mike Gannotti 🔳 retweeted
SpaceXSI : Grok Bot has been updated to version 0.75.0 Here are some of the new things they added: • See suggested prompts in an agent panel and dismiss the ones you skip. • Hover a library item to preview an image or video. • Get help when a Microsoft admin needs to approve a connection. Here are some things they have been working on but have not rolled out yet: • Control apps on your Mac from an agent. • Chat with teammates in a group around an agent. • Import memories from past conversations. • Invite teammates and open a permissions panel for what an agent can use. • Pick an avatar shape and color from a card in chat. • Connect a Microsoft Teams channel. • Hover a link in a message to see a preview card. • See analytics and a review queue for a team agent. Some other things that were worked on since the last update: • Open a setup flow to connect Glean as a remote tool. • Use tools on your Mac. • Open a details panel for a team bot, see who manages it and who can use it, and work through a review queue. • Sign in with access to plugins and stay signed in on your computer, or connect without an account. • Follow special rules for tools after stamps are reached. • Open a map and take actions in a workspace where bots can talk to other bots in the cloud, limited to your team. Note: features mentioned above may or may not be finished or active. The changes above are AI generated based on new flags and strings detected since the last release. The stable version could have differences. The official changelog does not list 0.75.0. For the official changelog see here: x.ai/changelog/bot
SpaceXSI : Grok Bot has been updated to version 0.72.1 Some other things that were worked on since the last update: • Add or remove a plugin from the marketplace, sign in when you add one, and connect it from the plugin details header. • Open a plugin’s details and connect from the header, read short app descriptions, and sign in when you add something. Note: features mentioned above may or may not be finished or active. The changes above are AI generated based on new flags and strings detected since the last release. The stable version could have differences. The official changelog does not list 0.72.1. For the official changelog see here: x.ai/changelog/bot
9
17
4
198
24,339
Took a peek over at Patricks progress on creating a dashboard on old Fire Tablet. Early in development but definitely making progress. With Grok Bot running on Omarchy we can fix everything
4
9
722