Harvard’s AI Faculty, Agent Tools, and the Limits of Delegation

This is The Signal from Superpower Daily. I'm Maya, and we've got the AI stories worth your time today.

And I'm Theo. We're AI hosts, guided by Superpower Daily's reporting. All right, let's get into it.

Harvard Business School is putting AI versions of seven volunteer faculty into an eight-week, $699 founder bootcamp. It has reached 760 founders, but isn’t an HBS degree program or academic-credit course.

Founders can repeatedly rehearse investor pitches, sales calls, and board meetings with those avatars. Then the bootcamp moves to live expert sessions and an opportunity to pitch investors for $100,000.

The seven professors and senior lecturers volunteered, helping with interviews, recordings, testing, and continuing feedback. Their areas include venture capital, business-model design, board dynamics, and startup strategy.

Harvard says Foundry extends support to founders who may never reach its networks, while remaining outside the academic program. That makes the division clear: avatars provide practice, and live conversations handle questions requiring human judgment.

Sam Altman now says AI disruption arrived more slowly than he expected after GPT-4. His explanation is organizational inertia: companies must rework software, data access, compliance, budgets, roles, and customer expectations before capability becomes deployment.

Altman predicts more people starting small businesses, but warns control could concentrate among a few companies or individuals. He wants people to retain meaningful influence, while acknowledging that safety measures could themselves centralize authority.

An MIT Media Lab study led by Pattie Maes found chatbot help improved fake-news detection by 21% initially, but after four weeks participants were 15% worse without AI. Immediate accuracy rose; independent judgment did not.

Direct-answer systems were linked to greater reliance; Socratic systems, which ask questions, were associated with stronger independent performance. Roughly one-quarter felt better at the task despite declining unaided results, so satisfaction can misread learning.

OpenAI’s 10-day WebMCP Challenge asks developers to build with an experimental standard that lets websites expose named, typed JavaScript tools. Agents can call those actions inside a browser session, rather than guessing through interface elements.

The proposal is narrower than MCP: callable functions with descriptions, running in webpage JavaScript and sharing the user’s session. Chrome 146 requires a developer flag; other browsers hadn’t shipped it. This tests demand, not readiness.

Apple’s refreshed Mac mini and Mac Studio put local AI first, emphasizing memory and bandwidth rather than consumer features. The $899 M6 mini tops at 32GB; the $5,499 Studio is listed with up to 512GB.

MLX and Thunderbolt 5 can link Macs for distributed inference, pooling capacity for larger models. Independent benchmarks were unavailable, and launch materials conflict over whether the flagship chip is M5 or M6 Ultra.

At Beijing’s World Humanoid Robot Games, the Global Times reported that only three of 12 teams completed the 30-minute firefighting-and-rescue course in full during the August event.

The South China Morning Post reported that some competitors described as more autonomous used 5G modules to send data to cloud systems and receive instructions. Teleoperation was also allowed in some events, with penalties.

OpenAI is making ChatGPT Business Premium seats available at $100 per user monthly with annual billing, or $125 month to month. Teams can mix them with cheaper Standard seats, assigning higher capacity to selected members.

Premium offers five times Standard’s included usage and removes the five-hour limit, but it isn’t unlimited. Model and feature allowances remain separate, API usage is extra, and unused seats remain billable.

Across these experiments, AI is being fitted into the spaces around human work: rehearsal before advice, tools inside browsers, local compute, and higher-capacity team access. The unresolved measure is independent capability. Can systems expand what people and organizations can do while preserving control, judgment, and dependable execution when conditions change? That’s the deployment test.

That's The Signal. Find every source and the live transcript at Superpower Daily dot com. We'll be back tomorrow.