Get tomorrow's brief in your inbox
Today: OpenAI launched ChatGPT app integrations with DoorDash, Spotify, and Uber, letting the assistant actually do things for you. Meta eyes layoffs affecting 20% of staff to fund AI infrastructure spending. Simon Willison shared how coding agents are making test-first TDD not just tolerable but essential.
ChatGPT Now Controls Your Apps (If You Let It)
OpenAI rolled out app integrations that let ChatGPT connect directly to your Spotify, DoorDash, Uber, Booking.com, Canva, and Angi accounts. Instead of just giving you advice, it can now take action: create playlists in Spotify, book hotels through Booking.com, design graphics in Canva, or match you with home service pros via Angi.
Setup is simple. Type the app name at the start of your prompt and ChatGPT walks you through signing in. Or go to Settings > Apps and Connectors to link everything at once. The convenience is real, but so is the data sharing. Connecting Spotify means ChatGPT sees your playlists, listening history, and personal info. Same pattern across other apps. OpenAI is betting you'll trade privacy for productivity, and for service businesses, that might be a fair trade if it saves 30 minutes of admin work daily.
The business angle: If you run a design shop, letting ChatGPT generate Canva drafts based on client briefs could cut first-draft time by half. If you manage travel logistics for a team, delegating hotel searches to ChatGPT could free you up for higher-value work. The cost is giving OpenAI visibility into your workflow data. Review permissions carefully before connecting business accounts.
LegalZoom's AI Hybrid Model
LegalZoom launched a human-plus-AI service that aims to cut the confusion and friction out of small business legal work. Instead of pure automation (which misses nuance) or pure lawyer time (which costs $300/hour), they're blending AI for routine tasks with human oversight for judgment calls. If you've ever abandoned a legal project because you couldn't tell if you needed a lawyer or a form, this targets exactly that gap.
No pricing announced yet, but the model suggests it'll land between DIY ($100-300) and full legal services ($2,000+). If you're choosing business structures, drafting contracts, or handling compliance, this could be the middle ground that actually works.
How Simon Willison Uses Coding Agents (Without Reading the Code)
Developer Simon Willison shared his framework for trusting AI-generated code in a fireside chat at Pragmatic Summit. His key insight: treat coding agents like external teams. You wouldn't audit every line from a vendor's API. You'd read the docs, test the integration, and trust professionals to deliver working code.
His workflow for every coding session: Start by telling the agent how to run tests (usually uv run pytest), then use red-green test-driven development (TDD). The agent writes failing tests first, then implements code to make them pass. Willison says he's hated test-first TDD his whole career because it felt slow and tedious. But with agents doing the work, it's free. And tests are no longer optional because the agent can spin through iterations without human frustration.
The tactical takeaway: If you're using Claude Code, Cursor, or similar tools, adopt this pattern. First prompt: "Here's how to run the tests. Use red-green TDD." Second prompt: your feature request. The agent will write tests, implement code, and iterate until tests pass. You review the test coverage and final behavior, not every line of implementation. Willison says Claude Opus 4.5 was the first model he trusted this way for classes of problems he's seen it handle before.
One more tip from the talk: agents can't test manually well, but you still need manual checks. Tell them to start the server and verify it boots. Just because the test suite passes doesn't mean the web server starts correctly.
Read more on Simon Willison's blog
Meta Eyes 20% Layoffs to Fund AI Spending
Meta is reportedly considering layoffs affecting 20% or more of its workforce (around 79,000 employees as of December 2024). Reuters reports this would help offset aggressive AI infrastructure spending and AI-related acquisitions. Meta called the report "speculative," but the company has done this before: 11,000 jobs cut in November 2022, another 10,000 in March 2023.
What this means for your business: The "AI is automating jobs" narrative is back, but as OpenAI's Sam Altman noted, many of these cuts are "AI-washing" where executives use AI as cover for over-hiring during the pandemic. If you're building AI into your workflows, don't assume it's a headcount replacement. It's a productivity multiplier for people who know how to use it. The real risk is competitors using AI to do more with the same team while you're still doing everything manually.
GitHub's Spam Crisis Shuts Down Open Projects
Jazzband, an open-source collective with shared push access for members, is shutting down because AI-generated spam PRs and issues made their open membership model untenable. Developer Jannis Leidel wrote that the project was designed for a world where the worst case was someone accidentally merging the wrong PR. Now only 1 in 10 AI-generated PRs meets standards. GitHub's own response was a kill switch to disable pull requests entirely.
If you run or contribute to open-source projects, expect maintainer burnout to accelerate. If you rely on open-source libraries, consider sponsoring maintainers or contributing human review time. The AI spam problem is forcing projects to choose between openness and sustainability.
Read more on Simon Willison's blog
Anduril Lands $20B Army Contract
Defense tech startup Anduril signed a 10-year contract with the U.S. Army worth up to $20 billion. The deal consolidates over 120 separate procurement actions for hardware, software, infrastructure, and services. Anduril, co-founded by Palmer Luckey (Oculus founder), is building autonomous fighter jets, drones, and submarines. The company brought in around $2 billion in revenue last year and is reportedly raising a new round at a $60 billion valuation.
Business angle: Enterprise AI buyers should watch how the DoD handles software procurement at scale. The shift from 120 contracts to one enterprise agreement signals a consolidation trend. If you sell B2B software, expect larger customers to push for unified contracts across product lines.
Test-Driven Prompting for AI Coding
Steal Simon Willison's 5-token prompt for better AI code: "Use red-green TDD." That's it. Before giving the agent your feature request, tell it how to run tests and add those five words. The agent will write failing tests first, then implement code to pass them. You get better test coverage, working code, and confidence in the output.
Visa Stablecoin Cards Now in 100+ Countries
Visa and Bridge (owned by Stripe) expanded stablecoin-linked payment cards to over 100 countries. If you deal with international payments or cryptocurrency customers, you can now settle transactions using stablecoins at 175 million Visa merchant locations. Phantom and MetaMask already use Bridge-enabled Visa cards. This won't replace your bank account, but it's an option for customers who prefer crypto settlement with lower volatility than Bitcoin.
Read more on Small Business Trends
ChatGPT's app integrations signal the next phase: AI assistants that don't just advise but execute. For small businesses, the opportunity is delegating routine tasks (playlist creation, hotel booking, graphic drafts) to an assistant that actually completes them. The risk is data leakage if you connect business accounts without reviewing permissions. Test-driven AI coding is becoming standard practice for developers trusting agents to write production code. If you're not yet using AI coding tools, start with the "use red-green TDD" pattern and manual verification steps. Watch for more enterprise contract consolidation as large buyers follow the DoD's lead on unified AI procurement.