It's hard remembering to come over here to Bluesky. I still spend a lot of time on tech X because that's where I know when shit is broken and all the devs are going to go whine about it.

Pregame for tomorrow’s challenge! 😆 Y’all better be rooting for team short stack AKA the troll twins 👹👹 Thank you for the opportunity @codetv.dev AKA @jason.energy ❤️❤️ CC: @thebethanyann.bsky.social @sterlingchin.bsky.social

I don't think so! Team Hailing Frequencies Open is gonna take the DUB!

RoxyRodBeck@roxyrodbeck.bsky.social · 2w ago

Pregame for tomorrow’s challenge! 😆 Y’all better be rooting for team short stack AKA the troll twins 👹👹 Thank you for the opportunity @codetv.dev AKA @jason.energy ❤️❤️ CC: @thebethanyann.bsky.social @sterlingchin.bsky.social

Why didn't I think of this before?! I have an OpenCode session running on my gaming computer and need to get back up to my desk to do other work. I gave Codex the goal to monitor the open terminal running an experiment. Now it's looping, and giving me updates!

Come by the @inngest.com booth at AI Engineer World's Fair and grab your very own laptop hand that holds your laptop open when your agents are running! I only printed 120 of them, so grab them while they last!

Are you gonna be in town for the AI Engineer World's Fair next week in SF? I'm gonna be wandering around the expo hall! Come find me at the @inngest.com booth or while I'm around the expo floor!

I love this! I gave codex a complex goal and it ran for over a day and completed the goal! Thank you OpenAI ❤️🫶

step.waitForEvent pauses execution until "post.approved" arrives, then resumes immediately. No polling loop, no cron job checking a DB flag. Set a timeout and it auto-cancels if approval never comes.

Scheduled and published by my own AI chief of staff. No Buffer, no third-party scheduler, just a durable Inngest function that slept until the right moment and posted. The agents are doing real work now. 🤖

{AI} in Production is back. July 2nd at the Inngest office in SF. I'm really excited to host this amazing group of speakers! Four talks from teams running AI in production at scale: 🔧 Cursor 🔐 Arcade 🎙️ Vapi ⚙️ @inngest.com Real demos, real production lessons, no vendor pitches. RSVP link in 🧵

It's here! The official @inngest.com Codex plugin! Built with all the love an attention of a builder who appreciates solutions to problems and making agent infra easier than before. Blog coming soon! github.com/inngest/inng...

GitHub - inngest/inngest-codex-plugin: Official Inngest plugin for Codex: audit codebases and build durable workflows and agents with Inngest.

Official Inngest plugin for Codex: audit codebases and build durable workflows and agents with Inngest. - inngest/inngest-codex-plugin

github.com

The recent tech layoffs are about freeing up capital to test AI?! Over the weekend, I was at a kids birthday party, and had a 4 hour convo with another dad. He works at <redacted>, and formerly at <redacted>, and asked me about my opinions on AI, agents I've built, and my predictions for the future

Engineer at a 1,000+ org, 2026 benchmark: "Retries, duplicate events, partial failures, long-running jobs... still hard to build systems easy to debug, resume safely, and trust at scale." Resources don't fix this. Connected stacks do. Link in thread ↓

The 19% of AI teams confident their stack scales 2-3x aren't using different tools. They use tools that are tightly connected. Show me what ran. Why it stopped. If it's safe to resume. All in one place. The work for 2026 is connection. Link in thread ↓

AX is the most important thing your company can be focusing on right now. How does an agent interact with your product. Your infrastructure and APIS are the most important commodities in the AX world.

Say it with me: AGENT EXPERIENCE Michael from @workos.bsky.social took the stage at MCP Night and talked about building systems that agents can use And your agents need durable infrastructure. That's what we're building at @inngest.com!!! #AX #AgentExperience #MCPNight

69% of AI teams in production are not using a third-party agent framework. They're calling LLMs directly or building their own. Top complaints: abstractions make failures harder to trace (26%), poor long-running support (19%), don't compose with orchestration (12%). Link in thread ↓

35% of teams running AI in production are not doing AI evals at all. Of the 65% who do, most built their own pipeline. 47% cite the same gap: eval state and orchestration state can't see each other. That's not an evals problem. It's a context problem. Link in thread ↓

I have a robotic arm with a camera in my office. I left Codex to its own devices. it can control the arm, look around, comment on what's happening, and it's focused on my cat. And my cat keeps looking at the arm. lol The infra was built with @inngest.com and the python sdk.

Sometimes AI vampires need to touch grass. Thank God for mobile hotspots.

"Debugging feels more like forensics than engineering." Engineer quote from our 2026 AI report. Observability = #1 unsolved problem (19% of responses). Orch-native observability: 40% diagnose in minutes. APM-only: 11x more slow-or-blind cases. Link in thread ↓

AI teams spend 2x the time on reliability vs non-AI teams. We split that by orchestration tool. 12-month change in burden: - Inngest: -10pp - Custom: -9pp - Temporal: +22pp - AWS/Vercel/CF: +9pp Tooling is bending that curve. The question is which way. Link in thread ↓

More engineers should mean more confidence. Right? We surveyed 130 engineers for the 2026 AI in Production benchmark. 26% of small teams are very confident at 2-3x scale. 13% of mid teams. 0% of teams with 500+ engineers. The gap isn't tooling. It's context across the stack. Link in thread ↓

Cron is older than I am. Ken Thompson wrote it in 1975. We still hand it our most important jobs. Now AI agents. Server down at 9am? Missed. Halfway failure? Stranded. Inngest gives cron durable execution. Same syntax. Recovers from anything.

5 things a platform owes an AI agent mid-task: 1. State that survives failure 2. Deterministic retry 3. Pause and resume across time 4. Failure as first-class 5. Multi-agent coordination Link in thread ↓

I'm testing an AI clone using HeyGen and ElevenLabs It's not that I want to replace myself, but there is a lot of content that I want to get out, and there isn't enough time to record and edit all the videos. I'm thinking of this as part of a learning program for new users. What do you think?

AGI needs more than scale. Demis Hassabis says we still lack continual learning, memory, and steady reasoning. Agents are the path, and small models plus distillation already let engineers do 1000x the work. If AGI lands near 2030, are you building with it in mind or ignoring it?

Weida Tan, co-founder of Press Club, speaking at AI in Production on May 6. Talk: "From Vibe Coding Demo to AGI-Proof Software." AI made impressive demos easy. Knowing what will last is hard. Weida's covering how to ship prototypes that survive the AGI tide. Inngest HQ SF, 5:30. Link in thread ↓

Google dropped DESIGN.md It gives coding agents a structured design system in one markdown file You define colors, typography, spacing, and components, then lint and export with a CLI If you build with AI agents, your design tokens need a spec like this Sources in 🧵

Matt Biilmann named four pillars of Agent Experience in Jan 2025: Access. Context. Tools. Orchestration. Pillars 1-3 have llms.txt, MCP, agents.json. Pillar 4 has: nothing. That's why I joined Inngest.