Maker Meetings

Most of an agent transcript is the agent getting its bearings — directory listings, file reads, edits that flubbed three times before they worked. The signal is the intent underneath: what the human asked for and what the agent decided. https://withstoa.com lets you and a co-founder prototype a tool to strip away the rest — live, in a shared agent sandbox. In this Maker Meeting, Jake works with Sean Johnson (CTO and co-founder of SpecStory) on a real internal problem: SpecStory histories are long, and feeding full transcripts into their PR-summary tooling burns tokens. They set out to build a compressor that distills a history down to the minimum representation of intent — keeping the human/agent dialogue and the decisions, dropping the mechanical tool activity. They turn the meeting's own transcript into a spec, then race two implementations head-to-head: Claude Code (Opus 4.8) building locally and Codex (GPT-5.5) building in parallel, with a third agent judging the results. Along the way they meet the "Ben Distiller," watch the agents discover and shame each other, and land on a surprising disagreement over how to treat context compaction. In this episode, we cover: 0:00 The problem: compressing Spec Story histories for PR summaries 1:11 What counts as "core context" — and why this is lossy compression 1:48 Preserve the intent, drop the tool-call details 2:18 Turning the transcript into a spec: meet the "Ben Distiller" 3:19 The rule of thumb: keep decisions, drop the agent getting its bearings 3:57 Going local-first with adversarial agents 4:55 Claude Code vs. Codex: a head-to-head build 6:13 Judging the two outputs — and getting "shamed" 6:43 The compaction surprise and the takeaway Follow: YouTube ▶ https://www.youtube.com/@UCXgR8mzS-Ja3pvA8xsuV9Tg LinkedIn ▶ https://www.linkedin.com/company/specstory/ X ▶ https://x.com/specstoryai Website ▶ https://withstoa.com ------- Song: Tunetank - Whiskey Blues (No Copyright Music) Music provided by Tunetank. Free Download: https://bit.ly/48Rl83b Video Link: • Smooth Relaxed Jazz (Copyright Free Music)

When a coding agent keeps getting something wrong, the instinct is to prompt harder. The better move is to have it build you a tool — a jig — that makes the task easy for you and the agent both. https://withstoa.com lets you and a teammate do exactly that, live, in a shared agent sandbox. In this Maker Meeting, Jake sits down with Nate Borwick (computer science and design student building with the SpecStory team) on a real design problem: the hero animation on Stoa's Share Localhost landing page — a paper airplane that flies along a curved path — looked off, and coding agents are terrible at drawing a curve they can't see. Instead of fighting Claude pixel by pixel, Nate had it build a visual path editor: a dev-only tool hidden behind a URL parameter, where he drags control points and copies the result straight back to the agent. They rebuild that tool live inside Stoa using shared agent sandboxes, then look at a second jig — a "construction zone" bookmarklet that turns on-page comments into agent-ready context. In this episode, we cover: 0:00 Meet Nate and the Share Localhost hero animation 1:08 Why coding agents can't "see" a curve 1:52 "Nate built a jig": a dev-only path editor 2:56 Rebuilding the tool live in a shared Stoa sandbox 4:21 First look, then iterating on the editor 5:50 A second jig: a construction-zone comment bookmarklet 7:11 The takeaway: have the agent build your tools Read the companion post ▶ https://withstoa.com/blog/the-jig-is-back Follow: YouTube ▶ https://www.youtube.com/@UCXgR8mzS-Ja3pvA8xsuV9Tg LinkedIn ▶ https://www.linkedin.com/company/specstory/ X ▶ https://x.com/specstoryai Website ▶ https://withstoa.com

New teammates need two things on day one: a map of what exists, and a shortlist of what matters. Most onboarding docs give them only the first. https://withstoa.com lets you ask the agent for both in the same project space — and pulls next steps straight out of your meeting transcripts too. In this Maker Meeting, Jake works with Antaripa Saha (data scientist and AI memory expert) on a real onboarding problem: "Team Time" is the Stoa space we've been using for months — full of great context and equally full of junk. They prompt the Stoa agent to generate a clickable Mermaid diagram of the directory, iterate on it cross-agent, debate what makes a file worth keeping (recency vs. relatedness), and then ask the agent to pull next steps directly from the meeting's own transcript — context that lives where you work. In this episode, we cover: 0:00 Welcoming Antaripa and the new-teammate context problem 0:50 "Great context, also a bunch of junk" 1:24 Visual diagrams beat markdown summaries 2:11 Cross-agent prompts and "be explicit about saving context" 3:29 Iterating in place: context persists across turns 4:19 Recency vs. relevance: a two-factor archival rule 5:30 Pulling next steps from the meeting transcript itself Follow: YouTube ▶ https://www.youtube.com/@UCXgR8mzS-Ja3pvA8xsuV9Tg LinkedIn ▶ https://www.linkedin.com/company/specstory/ X ▶ https://x.com/specstoryai Website ▶ https://withstoa.com

The fork in enterprise sales: executives buy on strategic story, the buying committee buys on a working demo. Most teams pick one. https://withstoa.com lets you build both in the same conversation. In this Maker Meeting, Jake and Greg work with Tom Noser (founder of Fortune's Path, a healthcare product consultancy) on a live client problem: a healthcare tech company wants to prove its products are worth more as a unified suite than as individual products. In one sitting they go from blank slate to a clickable matching prototype and an executive pitch deck — two validation artifacts for two very different buyers. In this episode, we cover: 0:00 Welcoming Tom Noser and the unified-suite hypothesis 1:21 What the products do and Greg's "UberX for nurses" playback 2:05 "Something to react to" beats a blank slate 3:02 Two buyers, two needs: executives vs. the buying committee 4:47 Pivoting to slides for the strategic story 6:04 The fidelity tradeoff: vaporware vs. concrete Follow: YouTube ▶ https://www.youtube.com/@UCXgR8mzS-Ja3pvA8xsuV9Tg LinkedIn ▶ https://www.linkedin.com/company/specstory/ X ▶ https://x.com/specstoryai Website ▶ https://withstoa.com

The PRFAQ is dead. When agents compress implementation time to near zero, the bottleneck isn't building it's figuring out what to build. https://withstoa.com is purpose built for this! In Episode 2 of Maker Meetings, Jake and Greg draft a product landing page (the new PRFAQ) for a feature idea on unmint.dev: freshness timestamps for AI agents. They brainstorm with an agent in Stoa, hit on the DigitalOcean lesson about "last updated vs. last reviewed," and whittle a big block of ideas down to what actually ships. In this episode, we cover: 0:00 Introduction to PRFAQs 0:50 Brainstorming new features for Unmint 2:10 The freshness idea: timestamps for AI agents 3:00 The sculptor metaphor: working with agents to whittle down 4:10 Reviewing the agent's draft and keeping humans in the loop 5:15 Whittling the feature list and wrap-up Follow: YouTube ▶ https://www.youtube.com/@UCXgR8mzS-Ja3pvA8xsuV9Tg LinkedIn ▶ https://www.linkedin.com/company/specstory/ X ▶ https://x.com/specstoryai Website ▶ https://withstoa.com

In the first-ever Maker Meeting, Jake (ex-CPO Docker), Greg (ex-CPO Pluralsight), and Isaac (ex-Answer AI) get together in Stoa (https://withstoa.com) to build a Calendly clone from scratch. Watch how three product minds plus AI agents go from blank canvas to working scheduling app live, in one session. This is the pilot episode of Maker Meetings, where we show how product work can actually get done in the AI era: real decisions, real tradeoffs, and real software shipped in a multiplayer room. Follow: YouTube ▶ https://www.youtube.com/@UCXgR8mzS-Ja3pvA8xsuV9Tg LinkedIn ▶ https://www.linkedin.com/company/specstory/ X ▶ https://x.com/specstoryai Website ▶ https://withstoa.com

In Episode 1 of Maker Meetings, Jake and Greg walk through a real product planning exercise using Ash Maurya's Lean Canvas framework. Applied live to unmint.dev, their open source, NPM-packaged documentation template on GitHub. Along the way, they use Stoa (https://withstoa.com) to turn the conversation itself into an updated canvas, demonstrating how multiplayer AI tooling can compress the loop from discussion to working artifact. Timestamps: 0:00 Introduction and why docs matter for startups 1:35 Applying the Lean Canvas framework 2:45 Customer segments and problems 4:25 Unique value proposition and solution 4:55 Handing off to the agent and key metrics Follow: YouTube ▶ https://www.youtube.com/@UCXgR8mzS-Ja3pvA8xsuV9Tg LinkedIn ▶ https://www.linkedin.com/company/specstory/ X ▶ https://x.com/specstoryai Website ▶ https://withstoa.com