John Paul Herold (@dailyherold) header
John Paul Herold (@dailyherold) profile picture

VP Engineering @ Sembi, formerly @SouthwestAir @ExamSoft. While: faith, family, futbol (ynwa), linux, 2ch audio

Dallas, TX
Joined February 2010
Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)Photo by John Paul Herold (@dailyherold)
When interviewing others, I don't forbid AI nor ignore it. I'd rather see how someone uses it to solve problems. Therefore, I've started asking candidates to share their agent session if they're open to it. And if they are, remind them to scrub it for sensitive data. Yay/nay?
1
23
For those who use Teams, your messages are valuable for helping create idiolect "models" (e.g. for yourself + agent writing) and to use as context for any general workflows. I created this teams-extractor skill (github.com/dailyherold/skill…) to help your agent read those messages, all local. Same pattern works for Slack of course and any high use chat. Other use cases: summarizing or catching up on channels, more powerful searching, finding themes across messages. The skill uses your browser's or native Teams app's local cache and leverages a helpful open source library for digging through the data effectively. I wanted something simple that didn't require a bunch of network calls and have found this quite helpful over the last few weeks. Disclaimer: "and remember kids", be smart about what model and plan you are using with potentially sensitive information. Follow your company's policy.
2
39
And yes, flame me for Teams use, trying to get things moving another direction. Gotta make today's reality better while I work on the next!
11
I've been wanting @GeminiApp to show me the top senders from my !inbox-zero gmail account since it was integrated into the webapp. Started working on gws powered skills when google workspace cli came out, and now finally know who is spamming me most! Huzzah! github.com/dailyherold/skill…
1
1
72
The googleworkspace cli (gws) is indeed pretty sweet. BUT, the generated skills (github.com/googleworkspace/c…) are pretty meh. Descriptions not triggering my agents effectively, not much guidance within the skills either. Plan on using them as reference to your agent for writing your own tailored skills. And leverage deterministic scripts for more advanced stuff you want your agent to do within your workspace.
1
41
Simple stuff, but having fun with skills and liking this agent peer-review approach: - I get to a point with agent A where I want agent B to take a look - "get this peer reviewed" - agent A generates a review "request" file full of context, lists all tmux panes and asks which pane to send-keys a nice prompt to (supports session:window.pane or unique pane ID) - agent B takes the prompt, reviews, and knows to put a response file next to the original request file - agent A is on standby to read whenever Why helpful? Sometimes I'm working with different models and wanting to sanity check. Or hand off initial research with a fast model to a more powerful model to review and take over. Or I'm in a session past optimal context utilization, and want to get a fresh agent with same model without compacting. Using it on plans, issues, and even ideas. github.com/dailyherold/skill…
1
24
To @dhh or others with similar setups, how are you passing back and forth between agents in different panes? Copy pasta? Peer review skill that has some guidance on persisting context somewhere and tmux send-keys to inject a prompt in diff pane?
13
AI fun: doing the things that you've wanted to do but didn't have the time to prioritize. My work moved me to a mac from my [preferred] linux laptop and for the last 6 months I just didn't dig in and figure out how to adapt my nixfiles repo to a darwin host. Lazy? Sure. Not wanting to invest time on mac because I don't care for it? Sure. But I had other stuff to do, the priority was low. Fast forward a few agent sessions and @claudeai + @opencode (@helicone_ai provider) were able to validate a multi-host config pattern, brought in nix-darwin, helped with isLinux/Darwin logic, etc. I even used the opportunity to look up my favorite tmux and vim config from my old ansible-based dotfiles repo and adapt to my 2026 nix config with suggested updates along the way.
1
30
2026: - Ghostty term - Moved back to tmux from zellij (was cool, but I have too much muscle memory with tmux to get me off, plus tmux seems to be supported by most agent harnesses) - Neovim - Claude, opencode, and @kirodotdev for work
1
56
Interested to hear more on ARA from @AWSOpen. The concept of a package registry spec for agents, skills, mcp servers is intriguing. Actively been thinking about how to manage skills as deps for my team, better ways to distribute, etc. github.com/ara-registry/spec
21
md wrangler is the new yaml developer
21
Been tracking this project after brainstorming ways of effectively testing that skill development for agents made a difference. Especially as the quality of LLMs continue to rapidly increase. Very cool to see the research that went into this and now the SkillsBench framework to help reason with skill effectiveness going forward. – at University Park, TX
Introducing SkillsBench, the first benchmarks that measures agent skills and how well agents use them. 86 tasks from 105 domain experts across 11 domains, every task is verifiable, human created and has verified Skills. SOTA model without skills score ~30% without skills. 🧵👇
1
2
326
Hanging with @gitlab at #GitLabTranscend. The last 3 weeks has shown the velocity of change in technology has crossed a new threshold. Intrigued how Gitlab is looking to go from human origin sequential SDLC stages to continuous development parallelized with agents – at Dallas, TX
1
82
Lot of talk on the amount of high quality context from repos, MRs, pipelines, issues, metadata, etc and relationships available to Agents through gitlab-org.gitlab.io/rust/kn…
46
Predictably, following my hype tweet, I get to enjoy a "proper measured view that is grounded in reality".
Wrote up about my personal journey from AI skeptic to someone who finds a lot of value in it daily. My goal is to share a more measured approach to finding value in AI rather than the typical overly dramatic, hyped bait out there. mitchellh.com/writing/my-ai-…
59
The last two weeks of AI dev has been wild, and seems like a new reality of compounding velocity. @openclaw [and @moltbook], @CloudflareDev, @claudeai, @OpenAI, all flying. Much of the development community, flying. It makes me feel like I'm "chasing the wind", a phrase I typically use for unwise pursuit. But, in this case I'm seeing those who chase get experiential practice at new ways of working and shipping value. Regardless of how fast tooling and patterns are changing at the frontier. And that seems much better than sitting in a chair thinking the wind will move you. Or wondering why the wind is closing doors around you. Therefore: chase, experiment, experience, explore. Let the agents cook. And share some recipes!
1
52