The AI Friday Brief
Good Tuesday, NOLA. May 26th brings a reality check after last week's agent hype. We're seeing honest takes on what AI actually does to your coding workflow, plus a fascinating Vatican moment that's surprisingly insightful about power and tech. Also tracking: real costs hitting big companies, and a thoughtful podcast on why agents still need humans in the loop.
Practical Takes on AI & Work
Using AI to Write Better Code More Slowly
Nolan Lawson walks through a genuinely useful mental model: AI makes you write *better* code, but often slower. The tool changes *how* you think about problems—you're more likely to refactor, test edge cases, and ask harder questions. This isn't a complaint; it's actually a feature if you care about code quality over velocity. Popular on HN—developers in the thread are sharing their own workflows.
Hacker News
Claude Is Not Your Architect. Stop Letting It Pretend.
A sharp critique of treating Claude as a senior design lead. The piece argues that AI models are better at implementation details than system architecture—they can optimize code but struggle with long-term tradeoffs and business constraints. Worth reading if you're delegating design work to an agent.
HollandTech (via previous brief)
The Cost Reality Check
Uber Says Its AI Spending Is Getting Harder to Justify
After burning through its annual AI budget in just four months, Uber is pumping the brakes. This is a candid moment: massive spend on AI pilots, but measurable ROI remains fuzzy. It's a cautionary tale for any company mid-AI transformation—enthusiasm and tooling are cheap; proving business impact is hard.
The Verge
What ClickUp's Mass Layoff Tells Us About the Future of Work
ClickUp replaced hundreds of employees with thousands of AI agents in a single swing. The move is bold—and raises real questions about displacement vs. augmentation. Read it for the business strategy angle, not the doom narrative.
TechCrunch
The Pope, Power, and AI
Pope Leo XIV's Encyclical Isn't Really About AI
The Vatican dropped Magnifica Humanitas, the Pope's first encyclical, and it's a surprisingly sharp takedown of concentrated tech power. Not a technical document—it's about human dignity in systems designed by and for elites. The Verge's coverage and Simon Willison's notes both dig into why this matters beyond the headline.
TechCrunch + additional coverage
Tools & Releases
DeepSeek Reasonix: Native Coding Agent with Built-In Caching
DeepSeek released a coding-focused agent that caches intermediate reasoning steps, so you're not recomputing the same analysis twice. Faster inference, cheaper API calls. If you're building a code generation pipeline, this is worth a test.
Previous brief follow-up
Hugging Face Publishes Agent Glossary
A useful reference guide: what's a tool, an orchestrator, a harness? As agents become more central to workflows, having shared terminology helps. Bookmark this if you're scoping out agent architecture.
Hugging Face
Worth a Listen
Why Agents Still Need Humans
NLW explores the next phase of human-agent collaboration. Using Dan Shipper's After Automation essay, the episode argues that automation doesn't eliminate work—it creates *different* work. Great listen if you're thinking about redesigning workflows, not just automating them away.
AI Daily Brief (NLW)
How the Engineer Behind Claude Cowork Actually Uses Claude
Felix Rieseberg, the engineering lead for Claude Cowork, talks through his actual workflow—not the polished demo version. Practical details on how Anthropic's own team uses the tool they built. The meta-angle is worth it.
How I AI (Claire Vo)
Also
- Local LLMs Perform Better When You Teach Them to Ask Questions First — A practical prompt engineering tip: asking the model to clarify ambiguous requests before answering actually improves output quality on local models.
- Everyone Is Navigating AI Security in Real Time — Even Google — Security isn't solved. Even the biggest companies are still figuring out attack surfaces and mitigations as they ship.
- A Reality Check on the AI Jobs Hysteria — MIT Tech Review steps back from the doomscroll: aggregate employment in developed countries remains stable despite AI hype. Worth reading if you're tired of both extremes.
- It's Time to Address the Looming Crisis in Entry-Level Work — The real issue isn't mass unemployment—it's the squeeze on junior roles that used to be training grounds. A more nuanced take on the jobs conversation.
- The 4 AI Team Members Execs Should Hire Right Now — NLW and Nufar Gaspar break down the actual roles—prompt engineer, data specialist, etc.—that leaders should be thinking about building.
- Microsoft Copilot Cowork Exfiltrates Files — Security vulnerability report: Copilot Cowork's file handling has issues. Relevant if you're evaluating enterprise AI tooling.
- How a 5x Founder Runs His Startup Solo With AI Agents — Ryan Carson demos his actual agent workflows for OpenClaw, Codex, and Devin. Less theory, more practice.
- OpenAI Partners With Grupo Folha and Grupo UOL for Brazilian Journalism — ChatGPT now integrates trusted Brazilian news with attribution. Another data partnership, but the transparency angle is worth watching.
- Datasette 1.0a30: Customizable Jump-to Menu and Agent Integration — Simon Willison's data tool gets agent-friendly upgrades. Useful if you're building data exploration interfaces.
- Issue Submissions Should Be in Your Own Voice, Not AI's — A reminder: using AI to write your GitHub issue or forum post loses signal. Your voice matters.
- Notes on Pope Leo XIV's Encyclical on AI — Simon's detailed breakdown of what the Vatican actually said and why it's more interesting than you'd think.
- CVE-2026-28952: Apple macOS 26.5 Kernel Vulnerability Discovered by Claude — Claude found a kernel vulnerability in the latest macOS. A small data point in the ongoing 'AI security researcher' narrative.
Pass it on