|
 |
|
Stop paying for steps your agent repeats
If you've built an agent with Claude Code, Hermes or Codex, the hard part was getting it to work. Making it cheaper and faster is a whole other project: tuning prompts, testing changes and hoping nothing breaks. Lagrange Cloud deploys the agent you already have and studies every run. After about five, it suggests a fix you can replay and approve before it ships.
The AIgent's take: An agent that works tends to stay exactly as you left it, because nobody wants to spend an afternoon reading run logs. I like the replay step here: a proposed change runs again on up to five recent inputs, Lagrange pays for those replays, and nothing reaches your callers until you promote it. Start with an agent that only reads, since replays run for real and anything that posts or merges will do it again.
Get free credits on us with code AIAGENT.
Deploy your agent →
In partnership with Lagrange Cloud.
|
|
|
 |
Today's drop
🛠️ Astryx Gives You Meta's Design System, Free
⭐ 22.4k · meta/astryxRepo
Meta just open-sourced Astryx, the design system that grew inside the company over eight years and now runs more than 13,000 of its apps. It's built so you and your coding agent work from the same components, and it's free under the MIT license. If you've rebuilt the same buttons and forms for the tenth time, start there.
A design system Meta built internally over eight years, now public and built for people and agents to design together.
You're either hand-rolling components for the tenth time or locked into a styling library that fights your agent's output. Astryx ships 150+ accessible components with brand-level theming and a CLI, and you and your coding agent use the same typed imports with no build plugin to configure.
Quick start: install Astryx, a theme, and its peer dependencies.
Catch: it's currently in beta, built on React 19+ and StyleX, so pin versions before you ship it to anything that matters.
Open the repo →
HubSpotAffiliate
HubSpot Marketing Automation, visual workflows that send the email, set the task or move the deal when a contact does something. See how HubSpot worksWe may earn a commission.
OpenMausBotTool⭐ 20.6k
A sidebar of real AI agents you talk to like contacts, each running its own model and tools. Catch: its README warns it has no token and no crypto tie, so ignore any coin using its name.
cinecodeRepo⭐ 2k
Describe a film and an AI agent codes every frame, sound included. Catch: it runs inside Claude Code or Cursor on your own machine, so you're trading convenience for a local agent setup you have to maintain.
UnstractTool⭐ 7.4k
Write one prompt and Unstract turns any messy PDF into clean, usable JSON. Catch: it's built for finance, insurance, healthcare and compliance workloads, so the setup assumes you have real documents to point it at, not a side project.
|
|
|
Stop Making Shoppers Spell It Out
|
|
“Running shos” shouldn’t return zero results. Neither should “red waterprof trail runners under $150.”
|
|
Traditional search needs exact keywords. Gladly understands what shoppers mean, typos and all, and turns dead-end searches into products that fit what they asked for.
|
|
See how Gladly reads between the typos
|
|
Turn Dead Ends Into Sales
|
 |
New here? Start with The Awakening. It's free, it takes one sitting, and it's where every reader should begin.
Take The Awakening →
Turn a messy incident into a write-up you can actually send someone
Start Here is one prompt from our pack, every issue, tested on ChatGPT and Claude. If you've ever had to explain what broke to someone who wasn't in the room, and ended up rewriting the same three paragraphs four times, this is the one. Tap Copy prompt, paste it into the AI app you already use, and fill in the bits in capitals there.
The line that makes it work: "Do not invent a cause, an owner or a deadline to fill the shape." Most write-ups fail because the writer pads a thin timeline with confident-sounding guesses. This line forces the model to say "not established" instead of inventing a root cause just to look thorough. That's what separates a document people trust from one they quietly stop reading.
TRY THIS: Open the chat app you already have on your phone right now, paste in the prompt, and drop in the last incident you had to explain to someone, even a small one, a missed deadline or a broken deploy. See how it handles the parts you don't actually know yet.
This is one of 250 in the AIgent prompt pack. The rest, plus guided builds with Mike on video, are inside the AIgent membership: Join the AIgent membership.
|
|
EARN FROM HOME
Get paid to train AIAffiliate
AI companies pay people to rate, fix and check what their models write. Open projects list pay of up to $17 an hour for AI evaluation and up to $125 an hour for specialist roles, US-based, part-time on your own schedule. Browse live roles and apply free.
See open AI training roles →
We earn a referral fee if you're hired through this link. It never changes your pay.
|
|
 |
|
All seven Anthropic co-founders plan to sell 3,378,378 shares each in the IPO. At the $74 midpoint of the price range that is just under $250 million apiece, and as a group they keep 51.1% of the votes. (TechCrunch)
The OpenSSH and musl login bug from yesterday's issue is now being exploited. Greynoise says a group broke into its honeypots within four days of disclosure, and OpenSSH has shipped a fix. If your agents run on Alpine or other musl-based images, patch today. (Ars Technica)
Meta launched Muse Ember, a small model for agent work priced like GPT-6 Luna and Claude Haiku 5.5 at $0.10 per million input tokens. Claude Fable 5.2, out this week, costs 80 times that. With three vendors at the same price, Simon Willison's point is the one to use: compare cost per finished task, not cost per token. (Simon Willison)
A former Meta engineer says AI wrote at least 90% of his code in his last months there, and it didn't even double his output. He also watched strong peers ship brittle, overcomplicated code. If you measure agent adoption by lines shipped, count the defects too. (Evan X. Merz)
|
|
Recommended reading
If you like The AIgent, a small group of operator-tier publications worth your inbox: see the shortlist.
|
|
Which section do you open this issue for?
|
|
|
 |
|
That Meta repo sat inside one company for eight years before anyone outside got to touch it. Worth asking what's sitting inside your own stack right now that would help someone else if you just shipped it. Reply and tell me what you're building with today's drops, I read every one.
|
|
See you Thursday.
|
|
Haven't taken The Awakening yet? It's free and it's where we'd start. Take The Awakening →
|
|
|
|
