←
Corey J. Gallon2,243 posts
Corey J. Gallon avatar

Corey J. Gallon

@CoreyGallon

Sharing insights from the frontiers of AI Engineering. 🇦🇺 Technologist. Investor. Coffee Nerd.

◷ Joined December 2008▧ 899 Photos and videos
Corey J. Gallon avatar
Corey J. Gallon@CoreyGallon·Nov 16

"Can I have the code that solved the CAPTCHAs?" Since my talk at @aiDotEngineer World's Fair, that's the question I get most. I've got something better. The Dark Arts of Web Automation is now a course. Building AI Agents for the Web is four weeks, live. You build your own agent, for the web apps you actually use: Week 1: it clicks and types the way you do, and checks that each step really happened. Week 2: it skips the buttons and uses the app's own hidden API. Week 3: it gets past pages built to stop it, including the four CAPTCHAs from the talk. Week 4: the site changes, and your agent repairs itself. Four live lessons, and office hours with me every week, on your project. If you already build with a coding agent, you're ready. The founding cohort starts December 2. Before then, I'm teaching one lesson free and live: how to find the hidden API behind any web app. Tuesday, November 24, 7pm US Central. Register even if you can't make it, and I'll send you the recording and the Hidden API Placemat: the whole method on a single sheet. The course, and your seat at the free lesson: agentsfortheweb.com/?c=now-a…

1:33
htmx.org / CEO of HROK (same) avatar
⟳ Corey J. Gallon retweeted
htmx.org / CEO of HROK (same)@htmx_org·Nov 16

some big news! after a long & intensive search process i am very happy to announce @JensenHuang as the next CEO of htmx very excited to see the direction he takes hypermedia & memetic warfare please give him a warm welcome stronger together!

Post attachment
Jay Nagy avatar
⟳ Corey J. Gallon retweeted
Jay Nagy@JayNagy·Nov 14

I’ve worked that heatshield my entire adult life. Peak career moment today To go from little 7 tile patches to a full reusable ship. Absolute marathon I’m so grateful I got to do something useful, that I was given the chance I don’t really have many good memories from this experience. It was so difficult. It almost looks easy now, but none of it came easily Please do something amazing with this rocket. This tech is such a gift, a love letter to the species

Post attachment
Corey J. Gallon avatar
Corey J. Gallon@CoreyGallon·Nov 4

This is a *chef’s kiss* salutation from the app in the morning. Nicely done, Claude team! 👌

Post attachment
Corey J. Gallon avatar
Corey J. Gallon@CoreyGallon·Nov 3

One of my agents is working a device support issue with a live customer service agent on my behalf. I've got it sending me notifications when it needs things that only I can provide (e.g. the code sent to my mobile phone to verify my account). It also prepared really, really good documentation (photos, annotations, logs, etc...) on the issues. It's saved me 4+ hours that this would have taken me to troubleshoot and document manually, plus the headache of sitting in a support queue and then the distraction of dealing with the customer support rep in real-time. This was trivially easy to do. I logged into my account using chrome-agent -- an open source tool I've built for just this sort of agent web usage. Then, I let my agent (Claude Code, in this case) troubleshoot the issue using both the web application and my local network to connect to the device. When it determined that it was an issue we couldn't resolve, it used chrome-agent to start a live chat with the vendor's customer support. This is so cool -- we live in the future! Feeling the AGI ...

Post attachment
Corey J. Gallon avatar
Corey J. Gallon@CoreyGallon·Nov 3

This is particularly useful for agents that you want to update with new information or additional tasks during their run.

Want your Claude Code sessions to talk to each other? Just ask. Type something like "Let @api-worker know the schema migration finished" (typing @ lets you pick a session by name) or "Ask the session in my other terminal whether the migration finished," and Claude writes and sends the message itself. There's nothing to turn on. It's on by default from Claude Code v2.1.224. Use /rename to give your sessions names that are easy to refer to, and /peers to see which ones a session can reach. I made a one-page explainer of how it works (attached). The short version: Who can a session reach? Its own subagents, agent-team teammates (agent teams are an experimental opt-in), and your other sessions on the same machine. If the session is connected to Remote Control, it can also reach your sessions on other machines and in the cloud. Messages between sessions on the same machine go over a local socket and never leave the machine. How does a message move? 1. Discover: list the live peers. A session's name is its address. 2. Send: make the first line stand on its own, because the receiving session shows it as a one-line preview until it's expanded. 3. Queue: the message waits in the receiver's queue. 4. Deliver: a busy session reads it between tool calls. An idle session starts a new turn. 5. Reply: send back to the address the message came from. What keeps it safe? - The receiver chooses whether to accept, hold or refuse incoming messages. - A message is information, not an order. It can't approve a permission prompt, change config or run commands. - No "permission laundering": a session shouldn't ask a peer to do something its own user blocked. - A successful send means the message was delivered, not that it was read. Agents can coordinate with each other, and each one's human still decides what it's allowed to do.

Post attachment