Yesterday's Top Launches: 2 Tools from October 16, 2026
Yesterday's two launches both address the gap between fast-moving agent work and the records meant to verify it, with Tiden checking code against intent outside the generating session and the other capturing on-screen details that never got written down.
Yesterday’s launch list was short but unusually coherent. Both entries go after the same gap from different sides: the work is moving faster than the record of the work. One checks that the code your agent just wrote matches what you asked for. The other watches your screen so the details you never typed down don’t vanish. As far as new developer tools go, that’s a narrow theme for a single day, and a pointed one.
Tiden
Tiden is an intent verification platform for coding agents, and its argument starts with a claim that’s hard to argue with once you hear it: tests that run inside the session that generated the code agree with that code by construction. A green suite only proves the agent agreed with itself. Tiden moves the verdict outside the session. Specs live beside the code as requirements that branch and merge like the code does, and every change gets checked against that record. Tests run in your own environment. No model grades the diff.
The mechanics are more concrete than the pitch. On first connect, Tiden reads the repo and writes down each spec, then flags the ones no test proves. The site’s sample snapshot shows web at 1,204 TypeScript files, api at 2,318 Go files, billing at 396, plus 312 linked Jira issues, with specs sorted under areas like Web App, Public API, Billing and Auth. One example spec, TIDEN-377 “Billing tier caps volume,” sits there marked with no test. That’s a useful thing to see about your own codebase, even before any agent touches it.
Sessions are where it gets practical. An agent working an intent called rate-limit-public-api moves through Started, Coding, Settle and Close, carrying a list of promises. Two get proven, one stays open, and the interface refuses to close the session. You can’t quietly ship half a feature.
Merges that skip Tiden still trigger a spec draft. The example diff replaces a TODO in api/public/v1/limiter.go with an Allow function, and beside it the spec moves from “Unlimited for all tiers” to the actual tier caps and Retry-After behaviour. Branch reviews list what was promised, what proved it, and what’s still open, with three choices at merge time: merge, ask for a test, or accept the risk. Release reporting gives one verdict per release, part by part — 2.6.0 in production at 205 of 205 coverage and 12 of 12 components — and production errors trace back to the release that introduced them.
Two honest notes. The demo numbers are tidily round, and mapping an existing codebase is the part I’d want to see survive contact with a messy monorepo rather than a clean sample. The product is paid, web-based, and connects to GitHub, Jira, Linear, Notion and Azure DevOps, with install through a shell script from tiden.ai. For teams already shipping agent-written code daily, the framing of intent as something versioned and proven is the part that sticks.
AirJelly
AirJelly is a free desktop agent that watches your screen and acts before you ask. Authorized context from Slack, Zoom, Google Docs, Calendar and email gets captured continuously, stored as a searchable timeline, and turned into tasks, briefs and reminders. It runs on Apple Silicon Macs and Windows x64, with Linux listed as coming.
The three-step loop is watch, remember, act. Watching means reading across the apps you already work in. Remembering means a full activity log plus a Memory feature you can query about anything from your past work. Acting means auto-captured tasks with due dates pulled from conversations and meetings, proactive alerts that fire at the right moment, and an end-of-day Daily Report.
The scenarios on the site are the kind that make the pitch land. A review confirmed in a chat message becomes a task with the conversation attached. Fourteen open tabs across papers and notes become a single research summary that includes a source the user had forgotten saving. An interview on the calendar thirty minutes out triggers a brief with background and talking points. It’s also positioned as a context supplier for Codex and Claude Code, which is a smart angle since those tools are only as good as what you paste into them.
Where I’d push back: continuous screen capture is a real ask, even with screenshots encrypted and stored on device, capture pausable, and per-app exclusions available. And the value scales with how much of your work visibly happens on screen. Anyone whose thinking happens on a whiteboard or in their head gets less out of it. The “second brain” label is also crowded ground; proactivity is the actual differentiator here, not memory.
Neither launch carried a community ranking yesterday, so there’s no vote leader to point at. On usefulness to a shipping team, Tiden had the sharper argument; on immediate try-ability, AirJelly wins because it’s free and installs in minutes.