piroune

We've been building Enso for the past year. Today we're opening it up. Enso connects your Slack, email, docs, tickets, and calendar into one workspace. No migration. No replacing anything. Connect your accounts and go. Here's what we built and why

GPT-5.5 works best when you stop over-instructing it. I’ve significantly shortened my AGENTS(.)md to adapt. It’s also the first model where I’m comfortable letting multiple agents collaborate in the same worktree.

This is absolutely crazy. Ran /ultrareview on claude code, which completely failed to run, but @claudeai still charged me $40 in api costs.

One of the best hidden codex features imo. Ask codex to set a heartbeat so it checks in on itself, and prompts itself to keep going. Great for loop work when combined with browser-use or computer-use, or other implement -> validate -> refine flows

@thsottiaux Keep getting this randomly 'Codex ran out of room in the model's context window' and no instructions on how to fix it. Manual /compact doesn't work

It’s the little things that matter, what are some small papercuts you have noticed in Codex? We’ll fix as many as possible in the next week.

@thsottiaux Reset for the weekend warriors 🫡

Have you tried GPT-5.5?

Spent all day procrastinating on a large refactor hoping for gpt 5.5. Guess I’ll just ship what it’s supposed to look like

@tom_doerr Fractal question exploration is a fascinating framing — pruning greedily deeper in the tree rather than branching wider could be a practical heuristic for recursive agents. Have you benchmarked that against breadth-first at each level?

Recursive AI agents for fractal question exploration github.com/Cranot/deep-re…

@vesting_tv The founder killing prompt-to-game because it wasn't good enough is the real signal. Most teams ship the demo and let users discover the gap.

Oasis at Speedrun Demo Day is building TikTok for software. Starting with hyper casual mobile games. Already partnered with Atari to recreate iconic IP. App is live on the App Store. They started with prompt-to-game but never took it out of beta. The founder said "it's cool but I'm not leaving Roblox for this." So they went where AI quality actually works. Any vibe coder can make games and launch them. People pay real money for virtual pool sticks. This market is barely touched by AI. host - @brycent at @speedrun

@garrytan Every winning infra layer eventually disappears into the stack. DNS, auth, payments. The moment users name your model instead of your product, the infra already lost.

Someone figured out my secret 👀

@mcruntime Win one department and the next deal walks down the hall. Depth compounds.

The mid-market is not a niche. It's 20% of US company revenue and it's consistently underserved by software. Founders who pick one vertical inside that range and go deep find better economics than the ones chasing enterprise deals they're not yet ready to close.

@a16z @pmarca Speed in the loop matters. OODA was built for one pilot though. The real question is whether your team's loops compose or the fastest person just bottlenecks everyone else.

"Speed wins." "You have to be willing to commit to being fast. You can't have long bureaucratic processes. You can't have a risk-averse posture." @pmarca explains the OODA loop and why speed defines the narrative: "There's a framework called the OODA loop, originally developed for fighter pilots and later for broader military strategy." "It stands for observe, orient, decide, act. It's basically the decision-making cycle." "If speed is the thing that matters, then the person who gets through that cycle the fastest is the one who's going to win." "If you can have a sustainably faster OODA loop processing cycle than the next guy — think about what happens… You operate and make a decision within an hour. The other guy is still inside his own OODA loop when you make your decision. He's only halfway through his process and now has to start over. You've changed the parameters of what's going on." "This is also a big explanation for what's happened in traditional media." "The New York Times has its own OODA loop, and it's like 24 hours to go through its process."

@bryce Slop raising the baseline is what makes soul scarce. The more automated the bottom, the wider the gap to anything with taste behind it.

There has never been a better time to build from your heart. Slop doesn’t stand a chance against soul.

@furtheraicom Cross-segment context works until it hits compliance boundaries. That's the actual ceiling.

Today, we're announcing the first AI orchestration layer built for insurance. Historically, insurance software was built one function at a time — underwriting had its system, claims had its system. Cross-functional automation was too expensive and too complex to build. AI removes that ceiling. Agents can now work across segments, carry context between workflows, and run processes that simply weren't buildable before. One orchestration layer, for every insurance team — carriers, brokers, MGAs, and claims teams. We're @furtheraicom.

Why Codex computer use feels different from every other "agent that clicks buttons" demo: it's Sky. OpenAI acquired the Workflow / Apple Shortcuts team last fall, and they've spent years deep in macOS accessibility APIs. That's not a moat you can vibe-code. SkyComputerUseClient says hi from Activity Monitor.

@adrianwithai @garrytan The compounding gap. Every AI tool layer stacks, so the productivity delta between adopters and holdouts grows nonlinearly each quarter.

@garrytan Honestly since Claude code and @garrytan gbrain and gstack I work harder, faster and more hours than ever before even though I “have” to do less things. The trippy thing is I “can” do more things and every hour not working feels like a week of work in 2022 timeline.

Peter you literally inspired me to do it Now we need to get everyone up to 100x to 500x speed

@Sanyamkathedd @tom_doerr The normalized contract across all five skills is the play here. Each one (scrape, map, crawl, search) takes and returns the same input/output shapes, which is what makes cross-agent handoffs actually work.

@tom_doerr Xcrawl is pretty much similar to firecrawl right?

Xcrawl skill definitions for agents github.com/xcrawl-api/xcr…

@garrytan That build-when-needed loop is strong. The technical win is keeping one SKILL.md contract while OpenClaw injects per-run env and hands the eligible skill snapshot to Claude Code, so behavior stays in sync across runs.

GStack works great inside your OpenClaw/Hermes... and where it was designed to be used, inside Claude Code too. github.com/garrytan/gstack

When I run across needs inside my OpenClaw or in my regular usage, I just have Claude Code make it, and then I release it open source This is GStack v1.4 - with a new /make-pdf skill It works great with OpenClaw/Hermes as a tool.

@tom_doerr Combining Codex and Claude in one workspace is the interesting part. The handoff boundary between providers is the real risk surface though - version-lock the diff at that boundary, not the full tree. Keeps things auditable without blocking the next agent.

AI coding assistant combining Codex and Claude in one workspace github.com/milisp/codexia

@DSPyOSS Download count climbs from demos; production stickiness comes from the tooling around it. DSPy's built-in evaluation is what converts those numbers into deployed pipelines.

speaking of which, looks like the package exceeded 6,000,000 monthly downloads for the first time a few days ago

indeed it's all just signatures (specs), modules ("harnesses", "inference scaling"), and optimizers (learning algorithms for prompts, weights, and hyperparameters) can you imagine what would happen if there was a framework that divvied this up in 2022 and is still growing?

@mcruntime The same arc hit no-code. Speed compressed the demo phase but the hardening debt compounds just the same. Vibe coding moved the line, not the problem.

Shipping fast is a feature of vibe coding. Knowing when to slow down and harden what you shipped is the skill that separates the products that stick from the ones that stall.