Thought #33 - Cautious Optimism and Quiet Launches

Hi lovely humans,

While the AI world is still catching its breath post-summer, this week’s news reminds us how embedded these tools are becoming - quietly rolling out in Slack, speeding up clinical trials, or handling chargebacks on your behalf.

That said, we're also seeing growing friction points: artists pushing back, consultants crossing lines, and universities still scrambling to define what good AI use looks like. And if you’re in a junior role, it’s worth noting how often those get quietly cut when AI tools ramp up.

As ever, there’s a balance to be struck. The best stories this week show AI slotting into specific, often under-the-radar use cases - not sweeping revolutions, but helpful evolutions.

Our Week in AI

Lovable Cloud

The vibe coding tool Lovable has had a big update (and team tbh feels this more than most, as we had just finished our content for a Lovable workshop, and then had to redo it all…)

Before you had to link Lovable to Supabase if you wanted anything backend or database related.

Now it’s all in the one place and beyond easy to use.

I’m still wowed by the fake to add a system for users to login you literally prompt:

Add login

It’s amazing - of course you do have to tweak from time to time. But for someone who can’t code or who wants to build a prototype quickly, this is a real leap forward.

AI New Releases

• Claude now available in Slack: Anthropic’s Claude can now live directly in your Slack workspace. For now, it’s best for summarising threads or helping write replies—though you’ll need the Pro plan for access.

• Coursera launches ChatGPT app: One for learners—Coursera is now accessible inside ChatGPT (via GPTs), offering direct help with course content.

• [Stripe Smart Disputes rolling out in November]: Stripe will automatically submit dispute evidence using AI (with a 30% success fee on wins). It's a no-win-no-fee setup, but automatic enrolment has raised some eyebrows.

• Google expands AI search mode: AI Overviews are now available in more languages and countries—part of Google’s ongoing shift toward generative search.

• Claude gets web search in API: A notable update for developers—Anthropic has added web fetch tools to its Claude API.

• And yes, it was OpenAI DevDay this week…
  We haven’t had time to dive into it all (team Taught by Humans is a bit buried in events), but here’s what stood out so far:

  • AgentKit launched: OpenAI introduced a framework to help developers build multi-step, tool-using AI agents. Early reports suggest it’s powerful but not the friendliest for non-developers—we’ll test it soon.

  • They want to be your OS: With updates across ChatGPT, APIs, and workflows, OpenAI is positioning itself less as a tool and more as a full-stack AI operating system.

  • Big numbers: They celebrated partner companies using over 1 trillion tokens in the last year—one of several signs of how embedded AI now is in enterprise infrastructure.

  • General sentiment? People seem quietly impressed. It’s less fireworks, more solid expansion.

AI News

Jobs and the Workplace

• AI isn’t replacing workers (yet): A new Yale study finds no evidence of widespread job losses from AI adoption so far. The nuance? It might be too early to tell.

• But AI-heavy firms are already cutting staff: King’s College London reports a 4% drop in overall headcount and a 6% reduction in junior roles in firms using AI intensively.

• 1 million jobs in London could change: A new (but hard-to-trace) report says AI could alter one million jobs in London alone. Specifics are thin, but the scale reflects real anxieties.

• AI in hiring: A new arms race?: The NYT highlights how companies use AI to screen CVs—while candidates now try to ‘game’ the system with their own prompts.

• Coventry University on AI skills in automotive: The skills gap is particularly wide in car manufacturing—where AI uptake is increasing, but upskilling hasn't kept pace.

Policy and Public Sector

• UK speeds up clinical trial approvals with AI: AI-supported reforms are helping slash trial timelines—great news for medicine (and less excitingly, government efficiency targets).

• AI Lab for UK policing: The Home Office is building an AI lab to explore ethical and practical uses of AI in law enforcement.

• Deloitte fined for AI-written report: In Australia, Deloitte must repay the government after using AI to generate a report without disclosure. A stark reminder: trust still needs transparency.

Culture, Crime and Caution

• Robin Williams’ daughter speaks out on AI videos: Zelda Williams has asked fans to stop sharing AI-generated videos of her late father—raising ongoing questions around consent, legacy, and likeness.

• AI scammers creating fake online shops: A warning from Trading Standards and police about increasingly convincing scam websites powered by generative AI.

• More people using AI to plan holidays: Whether it’s an itinerary or a hotel shortlist, AI is becoming a quiet co-pilot for travel plans.

• MrBeast is worried about AI clones: The YouTube megastar voices concern about generative AI creating deepfakes and lookalikes. It’s a niche risk now, but for digital creators, it’s looming.

• Reuters Institute: How the public uses AI for news: Most people using AI for news go to ChatGPT—not traditional media sites. Some trust it more than journalists, which is... a bit bleak.

Not Quite News, But Worth a Read (or Listen or Watch)

• Effective Context Engineering for AI Agents: A technical but useful read for anyone designing prompts or workflows—Anthropic explains how “context” is more than just a prompt.

• Domyn: a potential OpenAI challenger from Italy: Milan-based Domyn is hoping to become Europe’s answer to OpenAI. Whether they can overcome Italy’s strict laws is another question.

LinkedIn AI Poll

We are taking a week off. But would love some suggestions on what we should be asking people in the coming weeks?

Final Thoughts

As always we hope this was helpful!

Feel free to share this with anyone who might find it useful.

Next week, we will play with AI Agents Kit so you don’t have to.

Laura
Always learning