Why AI Still Needs Human Babysitting
AI can do 60% of engineering work. So why can't we trust it with 60% of the workload?
Understand the hidden reason most AI agents underperform and learn the context strategies that turn promising demos into dependable systems.
Today In Ai
1. OpenAI launches a Partner Network to fast-track enterprise AI adoption. The OpenAI Partner Network aims to certify 300,000 AI consultants by year-end to help companies scale faster. The announcement landed just days after 42 state attorneys general subpoenaed OpenAI over its advertising practices, data handling, treatment of minors, and model sycophancy. A major expansion push and a sweeping legal probe arriving in the same week is not a quiet run-up to an IPO.
2. AI-referred shoppers outspend regular visitors by 53% per visit. Consumers arriving via ChatGPT, Gemini, and other AI tools convert at a 54% higher rate, spend 53% more per visit, and browse more pages than non-AI traffic, according to Adobe Analytics data reported by Reuters. If you run a business with any online presence, AI traffic is no longer a rounding error.
3. Cybersecurity veterans push back on the Anthropic model ban. 76 security experts have signed an open letter arguing that the government export control on Claude Mythos and Fable leaves defenders without the strongest available tools. They also point out that the jailbreak that triggered the ban works on other frontier models too, making the targeted restriction look more like policy theater than a real fix. The group outlined 4 specific suggestions for regulating AI more effectively. Read the full letter.
From The Frontier
The headlines vs. the data. Tech layoffs hit 40,000 in May, the highest single month in two years. AI gets blamed constantly, but some analysts argue it's also a convenient cover story for overstaffed companies that were looking for an excuse anyway. Stanford's Digital Economy Lab is trying to cut through the noise with AI Economic Indicators, an ongoing evidence-based look at how AI is actually reshaping work.
What the numbers actually show. Stanford's data pulls from 25,000 firms and finds no dramatic surge or collapse in overall hiring since ChatGPT launched in 2022. The aggregate picture looks surprisingly stable. But that stability is doing a lot of averaging.
The junior squeeze. For workers aged 22 to 25, the picture is different. Employment in AI-exposed occupations has shrunk at 3.8% per year for this group, while the least-exposed roles are growing at 2.0%. Junior software developers and customer service workers are seeing the steepest declines. Less-exposed roles like home health aides have actually added junior workers.
Why this matters longer term. Stanford is clear these are early signals from a fixed sample. But if AI keeps shrinking the entry-level pipeline in knowledge work, the pool of future senior talent will shrink right along with it. Companies that hollowed out their junior ranks to cut costs may find themselves with a skill gap in four years that no AI tool can fill.
What people are actually watching and sharing
Organized chats. ChatGPT now lets you pin and organize your conversations however you like, a small but genuinely overdue update for anyone managing dozens of active threads at once.
Privacy update. One user noticed Anthropic quietly updated its privacy policy to include identity verification, a change worth tracking as AI labs face increasing pressure from regulators worldwide.
Americans vs. Europeans. Two viral AI-generated selfies compare American and European kids side by side and the comment section went deep on what it reveals about cultural bias in image models.
Maximum chaos. A Redditor asked ChatGPT to generate an image with the most ridiculous amount of content possible. The result is exactly what you would expect, and somehow still surprising.
Prompt Station
Most people use AI to fix individual slides. This prompt from Alex AI skips all that and builds the entire thing from scratch, title through closing call-to-action, with storytelling structure baked in. Drop it into ChatGPT, swap the topic, and you have a complete slide-by-slide framework ready to customize.
Build a full TED-style presentation in one prompt
COPY AND PASTE THIS PROMPT
Act as a world-class presentation creator. Create a complete, slide-by-slide presentation on the topic: [INSERT TOPIC]. Include title slides, key points, examples, data, metaphors, visual suggestions, and a closing call-to-action. Structure it like a TED-level presentation with storytelling flow.
Swap [INSERT TOPIC] with anything you need to present: a quarterly business review, a product pitch, a team training on AI adoption. For sharper output, add a second line specifying your audience, for example: "The audience is a board of non-technical executives" or "This is for a 10-minute all-hands meeting."
AI is in 60% of engineering work, but only 20% can be handed off without someone babysitting the output. Join this live webinar on June 24 (free) to see how top teams are using a context engine to level up.