The business model of digital advertising is in the midst of a transformation and trillions of dollars are on the line. In this week’s episode of our podcast, Radical Talks, Firsthand Co-Founder and Co-CEO, Michael Rubenstein discusses building the first Brand Agent Platform, transforming how brand advertisers and marketers connect with consumers in what is shaping up to be a complete replatforming of the advertising industry.
In conversation with Radical Partner Molly Welch, Michael argues that the current ad model is structurally broken and that fixing it requires a full rebuild, not incremental improvement. A new infrastructure is required that no longer caters to links and clicks, rather agent interactions, inference and tokens at massive scale. He unpacks how Firsthand’s Brand Agent Platform replaces static ads and fixed destinations with real-time, agent-driven interactions that adapt to each consumer’s in-moment needs, effectively delivering a personalized, real-time web. He explores why publishers can play a meaningful role in shaping a context-first ecosystem.
Listen to the episode on Apple Podcasts, Spotify, or YouTube.
AI News This Week
-
Why the A.I. Job Apocalypse (Probably) Won’t Happen (NYT)
Despite warnings of an AI-driven jobs apocalypse, macroeconomic data tells a different story. U.S. unemployment sits at 4.3%, average hourly earnings are stable, and demand for software engineers continues to grow even as tools like Claude Code mature. New York Times opinion columnist Ezra Klein argues that AI will follow the pattern of spreadsheets and computers, where falling costs unlocked latent demand and expanded employment in affected fields. As routine work becomes cheaper, the relational economy grows, with people placing greater value on human expertise and judgement.
-
White House Mulls Tighter Controls on Advanced AI (Politico)
Given the rapidly advancing capabilities of AI models, the Trump administration is considering regulation that would review frontier AI models. Pre-deployment vetting would resemble the FDA-style review process, with the U.S government already signing agreements with Microsoft, xAI and Google DeepMind to review their models ahead of release. The push gained urgency after Anthropic restricted access to its Claude Mythos model due to its cybersecurity capabilities, and after the UK AI Security Institute concluded that frontier offensive cyber capability is now doubling roughly every four months.
-
Why Private Equity is Making Deals with the AI Giants (Axios)
Frontier AI labs Anthropic and OpenAI have struck deals with mega private equity firms to launch enterprise AI services companies. Both ventures adopt the forward-deployed engineer model popularized by Palantir, embedding applied AI engineers directly with clients to build custom workflows. The arrangement accelerates AI deployment across the PE firms’ portfolio companies while giving the labs a faster path to enterprise adoption beyond what their own engineering teams can support.
-
A Blueprint for Using AI to Strengthen Democracy (MIT Technology Review)
AI is rapidly becoming the interface through which citizens form beliefs, exercise civic agency, and participate in collective governance. Authors Andrew Sorota and Josh Hendler compare the shift to the printing press and broadcast media, and point to early evidence that AI can strengthen democratic life. Field research shows AI-generated fact checks earn cross-partisan credibility that has eluded human efforts, and AI mediators are already helping citizens find common ground at scale. Several states and localities now use AI-mediated platforms to conduct public deliberation, pointing toward more responsive governance.
-
Research: Observability-Driven Automatic Evolution of Coding-Agent Harnesses (Fudan/PKU/Zhifeng)
As coding agents tackle longer-horizon software tasks, the surrounding harness of prompts, tools, and middleware increasingly determines performance, yet harness engineering remains a manual craft. Researchers have introduced Agentic Harness Engineering, a closed-loop system that exposes harness components as editable files, distills millions of trajectory tokens into structured evidence, and pairs every edit with a falsifiable prediction verified in the next round.
Radical Reads is edited by Ebin Tomy (Analyst, Radical Ventures)
