AI That Works is a monthly meetup for anyone building with AI in Austin: developers, founders, operators, go-to-market teams, and the AI-curious. Every session: live demos from people shipping for real use cases, a hands-on build, and honest conversation about what's actually producing results.
We don't care if you're writing code or mostly prompting. If you're putting AI to work on something real, you belong here.
This month
We'll build a self-improving research loop that turns your own Claude Code session logs into a map of what you actually use, and what you're missing. You point an Apify Actor at the session files sitting on your disk, and it extracts a structured usage taxonomy from every run: the tools you call, the MCP servers you connect, the packages you import, the domains you touch. Then it cross-references that against what you already have installed, so you leave with a list of the gaps in your tooling rather than a list of things you own.
If you run Claude Code daily, you'll leave knowing what your AI tooling actually looks like in aggregate. If you build with AI, you'll leave with the full stack: an Apify Actor that reads session transcripts, a scoring model that ranks the gaps, and a webhook loop that turns a finished run into research the moment it exists.
How the evening works
We start with why guessing at your own gaps doesn't work: you think you know which tools you lean on and which you're missing, but the only honest record is the session log at ~/.claude/projects/**/*.jsonl, and almost nobody reads it. Then we build the tool that reads it for you.
First, the capture layer. We build an Actor that parses Claude Code JSONL session files and turns raw transcript into a structured usage taxonomy - tool_use events, MCP server names, npm/pip packages, file extensions, keyword domains - so a messy log becomes queryable data.
Next, the gap layer. We cross-reference those signals against your installed skills and the plugin marketplace, then score each candidate with an adjacency formula so the highest-value gaps float to the top. A significance gate decides what's worth acting on, so a three-message session doesn't fire a research run and the loop never turns into noise.
Then, the automation layer. We attach an ad-hoc webhook at run start so ACTOR.RUN.SUCCEEDED fires with the run ID and dataset ID in the payload, and results get fetched the moment they exist, with no polling. We also walk through the four ways to wire Apify into an agent - skills, cron prefetch, webhook-triggered turns, and direct MCP - and what each one costs you in context window.
Finally, outputs that compound. We route findings into vault notes, skill scaffolds, and plugin stubs so the research becomes reusable instead of another tab you never reopen.
You'll leave with a working Actor in your Apify account and a repeatable pattern for an agent that reads its own history and improves on it.
Bring your laptop.
Agenda
-
5:30 pm - doors open and networking
-
5:45 pm - kickoff and presentation
-
6:00 pm - workshop
-
7:15 pm - wrap-up and chatting
-
7:30 pm - end
What you'll need for the hands-on portion
-
A laptop (macOS, Windows, or Linux)
-
An Apify account (free tier or higher) with enough platform usage left in the month to run workflows live
-
Claude Code installed, ideally with a few real sessions of history to analyze
-
Node.js installed
-
Your preferred code editor (Cursor, VS Code, or similar)
-
Some comfort with the terminal and basic JavaScript/TypeScript - you don't need to be an expert, but you'll be editing code during the build
Don't have all of this? No worries, you can follow along with someone at the event who does. The concepts apply regardless of your stack.
About the speaker
Jordaaan Hill is an Austin-based AI engineer focused on AI strategy and systems, helping teams turn AI tooling into workflows that actually ship.
He runs Organized AI, a community for AI-native engineers built around hands-on events and real builds, and is a regular fixture in Austin's AI builder scene. At this session he'll walk through the Apify Actor he built to read Claude Code session logs and turn them into a self-improving research loop.
About the host
Apify is the platform behind thousands of ready-to-use tools that give AI agents and apps access to web data, automation, and real-world information. But this meetup isn't a product pitch, it's a community. See what people are building, learn something new, and meet the people doing the work.
👾 Join the conversation on Discord.
Explore Apify at apify.com.
We will use your information to contact you about the event. For more information, see our Privacy Policy.
