AI Automation Digest #010

OpenAI closed its latest funding round (https://openai.com/index/accelerating-the-next-phase-ai/) at $122 billion in committed capital, valuing the company at $852 billion. The list of backers reads like a who's who of global finance — Amazon, NVIDIA, SoftBank, Microsoft, a16z, Sequoia, BlackRock, Temasek, and over $3 billion raised directly from retail investors through bank channels. That last part is new: regular people can now buy in through ARK Invest ETFs announced alongside the round.

The numbers they published are worth stopping on. One year after ChatGPT launched, OpenAI hit $1 billion in annual revenue. By the end of 2024, they were doing $1 billion per quarter. Now they're doing $2 billion per month. The company says it's growing revenue four times faster than Alphabet and Meta did at comparable stages, and they're on track to hit 1 billion weekly active users.

This is no longer a research lab raising money to keep the lights on. OpenAI is becoming the enterprise software layer for AI — 9 million paying business users, 2 million weekly Codex builders, and a product that's inside more company workflows every week. The $122 billion is about locking in infrastructure dominance before the next wave of adoption hits.

OpenAI acquired TBPN (https://openai.com/index/openai-acquires-tbpn/) — the daily live tech talk show The New York Times called "Silicon Valley's newest obsession" — and installed it inside the company's Strategy org. Co-founders Jordi Hays and John Coogan will stay, and OpenAI says editorial independence is explicitly protected in the agreement. The show can still be critical of OpenAI. Two weeks ago they bought Astral (Python dev tools). Now they own a media property with a direct line to the founders and investors who decide which AI platforms get adopted. Worth watching how the "editorial independence" promise holds up in six months.

Teams can now add Codex-only seats (https://openai.com/index/codex-flexible-pricing-for-teams/) to ChatGPT Business workspaces with no fixed monthly fee — you pay per token used. ChatGPT Business itself drops from $25 to $20 per seat annually. New workspaces get $100 in Codex credits per team member, capped at $500 per team. The number of Codex users inside Business and Enterprise accounts grew 6x since January. The pricing shift is clearly designed to get hesitant teams to run a pilot — at $0 fixed cost, there's no longer a budget approval reason not to start.

Google released Gemma 4 in its AI Edge Gallery iOS app (https://apps.apple.com/nl/app/google-ai-edge-gallery/id6749645337), and it was the top story on Hacker News over the weekend. The model runs entirely on-device — no internet required, nothing leaving the phone. It includes multi-turn chat, a "Thinking Mode" that shows reasoning steps, multimodal image analysis, and real-time audio transcription. For industries with strict data residency requirements — healthcare, finance, legal, government — on-device AI removes the primary objection to AI adoption. A phone that reasons through sensitive documents without touching a cloud server is a compliance team's answer to the thing they've been saying no to for two years.

GitHub COO Kyle Daigle shared internal numbers (https://simonwillison.net/2026/Apr/4/kyle-daigle/) that are hard to fully process: 1 billion commits in all of 2025, now 275 million per week. GitHub Actions jumped from 500 million minutes per week in 2023 to 2.1 billion this week. Daigle didn't say so directly, but the implication is clear — AI-assisted development is driving this, and it's not slowing down. Software is being written faster than at any point in history.

scan-for-secrets (https://github.com/simonw/scan-for-secrets) does one thing: you feed it your API keys, it scans a directory for any trace of them. It checks raw strings and encoded variants — base64, JSON-escaped, URL-encoded. Run it like this: uvx scan-for-secrets $OPENAI_API_KEY -d logs-to-publish/

Simon Willison built it because he keeps publishing Claude Code session transcripts and gets paranoid about accidentally leaking credentials. The problem is real — AI coding agents generate verbose logs, and a key that slips into context during a session can end up in a file you commit or share. You can set up a ~/.scan-for-secrets.conf.sh file with shell commands that echo all your secrets, so it always knows what to watch for across projects.

OpenAI's Head of Business Finance shared public data (https://simonwillison.net/2026/Apr/5/chengpeng-mou/) this week: ChatGPT receives about 2 million weekly messages about health insurance. Seven out of ten healthcare-related messages arrive outside clinic hours. Six hundred thousand of those weekly messages come from people in "hospital deserts" — areas more than 30 minutes from the nearest hospital.

This wasn't planned. Nobody at OpenAI designed ChatGPT as a substitute for primary care access. But when people can't get to a doctor, they look for answers somewhere. The implications cut two ways. For healthcare and insurance businesses: your users are already talking to AI about their health decisions. That's happening regardless of whether you've built for it. For anyone watching AI policy debates: the question is no longer whether AI is providing healthcare guidance — it's whether that guidance is any good.

"OpenAI was the fastest technology platform to reach 10 million users, the fastest to 100 million users, and soon the fastest to 1 billion weekly active users."

— OpenAI's funding announcement, March 31, 2026

That sentence is doing a lot of work. The word "soon" is the part worth thinking about.

Till next time,