Podwise Weekly Picks, Week 39: Agents Meet the Real World
Ten episodes from the week of September 21: workflow beats model, Robinhood's tokenised stocks, safe logs for agents, AI security, and the case for smaller models.
Weekly picks6 min read
The strongest argument on this week's shelf is unglamorous: the model is not the moat, the workflow is. It shows up in an enterprise-software founder's warning, in a database engineer's design for letting agents touch live data, and in a bio-security conversation about what happens when an agent can act on the world without asking first.
These are the ten episodes from the Podwise Weekly Picks shelf for the week of September 21. Show names link to the episode in Podwise.
The ten
Why AI cannot replace humans in enterprise
The Twenty Minute VC · 21 Sep · 71 min
UiPath's Daniel Dines on why work processes, not models, will be the valuable asset in enterprise AI, and what Europe keeps getting wrong about building companies. The contrarian energy is the point: this is someone selling automation arguing that automation has limits.
Robinhood and the SEC's innovation exemption
Bankless · 21 Sep · 38 min
Tokenised stocks, trading that happens outside market hours, and what changes when equities become programmable. A crypto-native view of a market-structure shift that most AI coverage ignores; useful context for agents that trade.
Forkable shared logs for agents on live data
TigerBeetle · 21 Sep
Ram Alagappan on why streaming systems built for predictable programs fail when the consumer is an autonomous agent, and how a forkable log lets an agent experiment without writing to production. The most concrete infrastructure idea of the week.
Jev: system-one models for production
Latent Space · 21 Sep · 141 min
Diogo Almeida, a co-author of InstructGPT, argues that API-available frontier models have been optimised down one path and that production needs something faster and narrower. Long, technical, and the best explanation of why "small model, specific job" keeps winning in practice.
Harvey's margin whiplash and human-powered agents
TBPN · 22 Sep · 135 min
A news show with range: legal-AI margins, human-in-the-loop agents, a rebrand, and Ben Thompson on what assistants do to consumer habits. Skip the segments you do not care about; the Harvey discussion is the one with numbers.
An Oscar, two asteroids, and the algorithm in your sklearn
Latent Space · 22 Sep · 121 min
John Platt on Google's Empirical Research Assistance, AI for science, and a career that includes an Academy Award and two textbook algorithms. The rare episode that is both a biography and a working explanation of why ML fundamentals still matter.
Bio-security is an AI arms race
Latent Space · 23 Sep · 92 min
After the OpenAI-to-Hugging-Face incident, Eric Nguyen on why biology is one of the two areas frontier labs flag, and what defensive capability looks like when the attacker is also a model. Uncomfortable, and worth an hour.
OpenRouter: from seed to Stripe
Latent Space · 25 Sep · 81 min
Alex Atallah and Anjney Midha on becoming the neutral routing layer for more than ten million developers, and why model diversity is the bet. The business-model episode this week; read it with the "beyond the god model" pick from week 40.
How do you secure a world of AI agents?
The a16z Show · 26 Sep · 56 min
Aaron Levie, Steven Sinofsky, and Martin Casado argue that the safety conversation is happening before we have defined the risks, and draw on earlier waves of computing to say what regulation should wait for. A good counterweight to the more alarmed episodes in this series.
Five predictions for a world of agents
The Twenty Minute VC · 26 Sep · 57 min
Parag Agrawal, now building search infrastructure for agents, on what breaks when agents query the web a thousand times more often, and why he thinks the advertising model does not survive it. A founder's predictions, so treat them as bets rather than forecasts.
How these are picked
The shelf is curated by Podwise from the week's episodes, and the order follows the collection. All ten are processed, so reading a summary or transcript costs no credits.
Browse the full shelf, including past weeks and themed collections, on Explore.
