You're being watched

gm legends, happy Friday.

An AI that watched how you work and turned it into software before you asked, email campaigns you describe instead of design, and a $5B video company that just eliminated the editing step. Plus: the forums asking whether everything is actually an agent.

Yansu is BentoM'sL logic pointed at one person. Bozhao Yu built BentoML to serve AI to millions of apps at scale. Yansu does the same for your workflow: it runs in the background, watches your screens and messages, finds the patterns in how you work, and builds custom apps from them before you think to ask.

🔥 Our Take: Every automation tool starts with you explaining your workflow. This one skips that. It watches your screens, reads your patterns, and builds the tools before you ask. If that sounds useful, you already know what you'd want built. If it sounds invasive, you've identified the real tradeoff.

Brew came out of 10,000 hours Philip Sørensen spent making email his highest-ROI channel at Revolut and Nova Credit. The product he built: describe the goal, the audience, and the sequence in plain English, and it generates the copy, design, audience segments, and automation logic all at once.

🔥 Our Take: The standard email tool workflow: pick a template, drag blocks, write copy, configure audience, set automations. You're building before you know if the idea is worth it. Describe the campaign first and let everything else generate, and you find out if the concept is sound before you've spent an hour in the builder.

Runway Agent takes a brief in chat and outputs a finished, edited, sound-designed video. No timeline, no effects stack, no audio sourcing.

🔥 Our Take: Runway already changed what it takes to generate a clip. This changes what it takes to finish one. If you've ever spent three hours in an editor sourcing music, cutting to beats, and mixing levels on a 30-second ad, that's the specific thing this is going after. The CEO said AI could help Hollywood make 50 films instead of one blockbuster. The Agent is what that looks like at the short-form end.

Amr Aniy (@amraniyasser) posted the question the PH community has been sitting on: not whether agents are real, but whether every product actually needs to be one. The thread is about the gap between the word and what most products are actually doing.

The replies get specific. One commenter put the ROI problem directly: firing engineers at $150K and spending $200K in tokens to do the same work. Another cited real numbers: without proper infrastructure, agent success rates sit at 23%. With error handling and the right tooling, 81%.

The sharpest take reframes it: "The real disruption isn't agents replacing apps. It's agents replacing interfaces."