Skip to content

Podwise Weekly Picks, Week 40: Personal Agents and the Price of Compute

Ten episodes from the week of September 28: an agent you can text, SAP on what AI does to software jobs, GPU depreciation, and the case against one model to rule them all.

Weekly picks6 min read

Podwise Weekly Picks, week 40

Personal agents stopped being a demo this week: one founder has an assistant with its own phone and computer, and another episode benchmarks the dozens of products trying to do the same. Meanwhile the infrastructure side asks the less exciting question — what is a GPU actually worth after two years?

These are the ten episodes from the Podwise Weekly Picks shelf for the week of September 28. Show names link to the episode in Podwise.

The ten

Building Instinct, the personal agent

Invest Like the Best · 28 Sep · 86 min

Noah Shinn built an assistant you text, call, or email like a person, with its own phone and computer. It is invite-only and growing about 10% day over day without marketing, which is either a signal or a small base. The interesting material is the behaviour people settle into when the agent can act for them.

Claude Code's next era

Latent Space · 29 Sep · 92 min

Thariq Shihipar on what Anthropic has shipped into Claude Code and what the product is becoming now that the model is no longer the constraint. A useful listen if you are deciding how much of your workflow to move into an editor agent.

The personal agent race is here

The a16z Show · 29 Sep · 51 min

Anish Acharya and David Pawlan, who built an assistant benchmark, on testing dozens of agents against real tasks: email, travel, financial admin. The claim worth considering is that the best agents will become invisible — no app, no interface, just a message thread.

SAP's CEO on AI and the future of software

Big Technology Podcast · 30 Sep · 61 min

Christian Klein argues AI will not kill software but will change the jobs around it, and that reliability is the thing that keeps an incumbent's moat intact. A rare episode from the enterprise-software side rather than the model side.

Why Dwarkesh is wrong about computer use

Latent Space · 30 Sep · 39 min

A corrective episode: what computer-use agents actually do well, where the critique lands, and how OpenAI shipped a competitor to a widely discussed demo in a week. Good for calibrating your expectations before you promise an agent to a customer.

Why Wall Street is wrong about the next Fed move

Bianco Research · 1 Oct

Jim Bianco thinks the AI boom is in a 1997–98 skepticism phase rather than a euphoric peak, and that the real danger is hyperscalers paying 8–10% on debt and crowding out ordinary businesses. Also a detour into diesel shortages and the war. A macro episode that explains why the AI trade keeps wobbling.

Frontier chips for frontier labs

No Priors · 2 Oct · 36 min

Fractile's Walter Goodwin on building a full-stack chip company when Nvidia, AMD, and Broadcom are already accelerating, and on co-designing silicon with model architecture. A short, technical counterpoint to the "just buy more GPUs" assumption.

Eric Schmidt on the future of war

The Economist · 2 Oct

Schmidt on drones, Ukraine's military-tech capability, and what a decade of defence investing has taught him. Not an AI episode in the usual sense, but the clearest example this week of a technology changing an institution faster than the institution can adapt.

The future of data centres

The Twenty Minute VC · 3 Oct · 61 min

Crusoe's Chase Lochmiller on why everyone gets GPU depreciation and energy costs wrong, who actually makes money from AI, and why most moats in the space do not exist. A good companion to the Bianco episode: one talks about the debt, the other about the asset behind it.

Beyond the god model

The a16z Show · 3 Oct · 48 min

OpenRouter's Alex Atallah and Replit's Amjad Masad argue the future is an ecosystem of specialised models routed by job, not one model to rule them all. If you are building a product on a single provider, this is the argument to answer.

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.

Podwise

Skip the six steps. Just listen.

Podwise transcribes, summarizes and maps any episode for you.