What I’m Focusing On Now#
Professional Life#
SVP, Engineering at Lark, March 2023 - May 2025.VP Engineering at Strive Health. Joined October 2025.- Chief Tentacle Officer, Groktopus LLC. Still alive, still writing about AI and the future of work over at groktop.us.
What Changed Since I Last Updated This Page#
October 21, 2026#
Work. I’ve been looking for my next role for several months. There have been a few close calls where I thought I’d found the right thing, but they didn’t work out as hoped. I’m in final talks with another organization this week and expect to know which way things are going by Friday. I’m optimistic. They seem invested, and so am I. What excites me is the combination: a strong mission focused on public good and the chance to double down on my AI work. Nothing is settled yet, but this one feels promising.
GroktoCrawl. The exciting part of the experimental branch was making research something you can keep working with, rather than an answer that disappears into a chat history. It keeps the source material behind a report, lets you inspect the exact passages it cites, and gives follow-up questions that evidence to build on instead of starting from scratch. You can also bring your own documents alongside web sources. That foundation is now in the main project, with live research in version 1.0.1.
Since that release, fixes have landed for comparisons that researched only one side, document citations that didn’t lead back to the source, and failures that didn’t explain what went wrong. The goal is research you can return to and check, not just a convincing-looking answer.
SlopSearX. I’m about two-thirds of the way through converting it from Python to Rust. The search service, research runner, and many of the connections to outside sources have already moved over, though there’s still integration and testing to finish. I’ve had to pause because the conversion is chewing through my OpenAI allowance, even on the $200 subscription tier. I’ve been running out of resets, too. Rust is less forgiving of the mistakes an AI coder wants to make. That slows things down, but I see it as a good thing. My hypothesis is that the result will be better software. I’m looking forward to testing that rather than declaring victory before it’s finished.
Agent Skills. The collection has grown in some useful directions beyond coding. Teach Me adapts explanations and practice to what the learner actually understands. Infographic helps turn evidence into clear visual explanations. Amazon Memo helps with narrative decision documents, and Meetings covers preparation, facilitation, accessibility, and follow-through. Existing skills have had attention, too: Product Discovery now works from a real draft rather than dragging you through a generic interview. The software-development guidance has stronger independent review, checks for missed obligations, and a way to turn failures into small, reviewed improvements. More skills are nice. Skills that do useful work without losing the plot are the point.
October 4, 2026#
SlopSearX 0.6.0 is out. It adds optional Jev specialist routing and result reranking with a deterministic fallback, groups scholarly works before reranking, and makes it clearer when an engine can’t honor a requested filter. It also ships native, tested arm64 images. The release notes list the rest.
September 30, 2026#
GroktoCrawl X is a separate experiment in web research. The original GroktoCrawl gathers web sources and turns them into research. In the X fork, claims are tied to the source material behind them, so readers can see where an answer came from and judge the evidence. X remains experimental; the original is still the main project.
Search and AI review are easier to judge. SlopSearX now makes exact matches easier to find and keeps the source link beside each result. It also distinguishes provider failures from a genuine empty search. In CodeReview, developers can see what the AI examined and what evidence supports a finding.
Memory repair and Firefox research are safer. Hermes Cashew now makes and verifies a backup before repairing a memory database. The original GroktoCrawl no longer aborts deep-research submissions in Firefox.
I’m learning video creation and think I’ll try starting a YouTube channel soon. I don’t know yet whether the videos belong on the blog. The public Agent Skills library now has a thumbnail workflow to help creators make images that match the video and remain readable at small size, plus a guide to clear questions and source-backed choices for decision-support tools.
A few fixes landed in projects other people maintain. Laya.cpp rejects oversized requests instead of silently truncating them, and clients can select the model the service advertises in a separate update. Upstream Cashew no longer reports a memory update complete before the data is saved. NC Meshnet added three regional networks to its directory. My AI tool for managing short links is easier to troubleshoot with clearer logs and metrics.
Open Source Projects#
I’m maintaining a few things right now:
hermes-cashew. A thought-graph memory provider for Hermes Agent. Gives AI agents persistent, structured memory across sessions instead of starting from zero every conversation. Apache 2.0, built in Python, getting real use.
hermes-SkillOpt. Controlled skill optimization for Hermes Agent, derived from Microsoft Research’s SkillOpt.
groktocrawl. Self-hosted web scraping and research stack. What started as a Firecrawl alternative turned into something more capable: multi-page crawling, JS-rendered content, LLM-friendly output. Runs on Docker, works with any agent framework.
SlopSearX. Cloud-native, stateless, AI-agent-first meta search engine. Drop-in SearXNG replacement for the GroktoCrawl stack.
artifact-pyramids. Progressive disclosure for what AI agents produce: Summary, Analysis Collection, and Dossiers. The Artifact Pyramid is a structured methodology for organizing AI agent research outputs across three layers of increasing depth. Just as progressive disclosure governs how we feed agents context, the Artifact Pyramid governs what they produce, enabling downstream agents and humans to consume at the depth they need.
hugo-cyberdelic. A cyberdelic Hugo theme: 1969 psychedelic concert poster meets cyberpunk holographic future.
agent-skills. A growing library of reusable skill modules for Hermes Agent. Reference patterns, workflow templates, and standalone skills I’ve extracted from real use. If you’re running Hermes, there’s probably something useful in here.
hermes-profiles. Curated Hermes Agent profiles for specialist swarms. Opinionated, Hermes-optimized role definitions for delegation, kanban pipelines, and multi-agent orchestration.
programming-principles. Programming principles distilled from 14 classic software engineering books, packaged as an Agent Skill. The lessons without the 400-page commitment.
hermes-claude-import. Import Claude.ai conversation exports into Hermes Agent memory providers. Community collaboration, open source.
hermes-council. A multi-agent debate system for Hermes Agent, now running as a native plugin with typed contracts and structured inference. Composes panels of expert personas to surface hidden assumptions and stress-test ideas. MIT licensed, open source from day one. I use it for everything from editorial review to strategic analysis.
What I’m Reading (or Rather, Listening To)#
I’ve just started the audiobook of Dune Messiah by Frank Herbert, the second book in the Dune series.
This is my now page, inspired by Derek Sivers. It’s a snapshot of what I’m focused on at this point in my life. I promise not to let another four months go by before the next update.
More research on my about Now Pages note.