Every experiment.
Failures included.

the bottleneck was never the coding

OpenAI's Symphony post made the agent future feel less like better prompting and more like a new way to organize work.

the hardest part is not building the product

This week reminded me that the hard part is not making the product work. It is making people get it right away.

my workday is turning into one long audio stream

I started exporting my dictations and realized I am not really taking notes anymore. I am working out loud.

I improved my app by 44% without writing a single new feature

I adapted Andrej Karpathy's AutoResearch loop and pointed it at my diarization error rate. Went from 34.1% to 19.2% — a 44% improvement. No new model, no new features. Just a loop that found the one knob that mattered.

the autoresearch loop is running. baseline is 34.1% DER.

I said I'd build the diarization AutoResearch loop. I built it. The loop is running right now. Baseline DER: 34.1%. Here's the full story including all the stuff that broke.

everyone's building the same personal agent

every.to just shipped plus one. the personal agent pattern is converging fast and i'm not sure anyone has the moat they think they do.

six agents run my content pipeline and most of them are broken

i have 6 claude agents running overnight. the competitor digest is great. the content batch can't save files. the idea collision engine can't find the vault. shipping is still manual.

what if my app fixed itself overnight

karpathy's autoresearch ran 700 experiments in 2 days. my app breaks with more than 2 speakers. what if i pointed that loop at my problem.

diarize might change the pipeline

Found a library claiming 7x faster speaker diarization than pyannote on CPU. Haven't tested it yet. But if it's real, Transcripted's pipeline changes.

notion wants your meeting notes

Notion shipped custom instructions for AI meeting notes plus background mode. When the gorilla enters your space, you have to think about what you actually are.

should have just rewritten it

rewrote draft's entire ui from swiftui to appkit. should have done it weeks ago.

someone built the thing i'm building

Found a Show HN that's basically Transcripted. Local transcription, Obsidian vault integration, Claude Code skills. Same niche, same tools.

built the machine that builds the machine

Six agents running overnight. Zero features shipped to the actual product.

i woke up to 4 merged PRs i didn't write

Scheduled Claude Code tasks ran overnight on Transcripted while I slept. Security audit, CLAUDE.md sync, refactor sweep, code review. Four PRs. All merged before I had coffee.

Paperclip was the red bar. Claude Code is the pivot.

I spent weeks trying to get a local agent system reliable. It wasn't. Now I'm trying something completely different.

v0.5.0: the pill got a personality

The floating pill is completely rebuilt. Onboarding is smarter. Model downloads actually tell you what's happening. And the nightly agents caught two security issues I would have missed.

The memory layer

I've been thinking about Transcripted wrong.

What Transcripted is actually for

I've been building the wrong version of this app.

Building the agent system

I spent most of today trying to get a local AI agent system running and breaking it in new ways every hour.

Transcripted v0.3.0

Eight PRs merged today. That number is misleading.

Transcripted v0.3.0: PyAnnote, simpler onboarding, UX cleanup

Shipped Transcripted v0.3.0 — replaced Sortformer with PyAnnote for offline diarization, cut onboarding from 6 screens to 3, cleaned up the floating pill and transcript UI, and fixed a batch of edge-case bugs.

Transcripted v0.4.0

Incremental improvements to things that shipped rough. The core is solid. The surface layer is catching up.

90 minutes of agentic coding chaos

A Claude Code session that started with 'install Draft' and ended with a feature I might scrap entirely.

Transcripted v0.2.0: aurora default, branding stripped

Shipped Transcripted v0.2.0 — aurora recording animation is now the only UI, removed idle branding for a cleaner pill, and the whole ML pipeline processes 11 minutes of audio in 5.4 seconds on M5.

Transcripted is live — free, open source, and yours

transcripted.app is live. Free meeting recorder for Mac. 100% local, no cloud, no bot in your meeting. Speaker diarization that learns your team.

r3d.bar goes live

Bought the domain. Connected Cloudflare. Deployed from GitHub. The whole thing took one Claude Code session. First green bar for the site itself.

Draft onboarding test: 12% completion rate

Ran 50 users through Draft's onboarding flow. 12% made it to their first generated message. The mic permission prompt is killing momentum.

Transcripted pricing: $0 conversions at $9/mo

Zero paid conversions in the first week at $9/month. The free tier is too good — nobody hits the paywall. Time to rethink the wedge.

Draft voice detection hits 94% accuracy

After 200+ accepted drafts from 10 beta users, Draft's style matching hit 94% accuracy. The RLHF loop is working. Users are editing less over time.

CLIs are the future of AI integration

GUIs were always a compromise. Now that AI agents are real, the command line is back — not because developers are nostalgic, but because it's the only interface that actually works for AI.