Latent Space: The AI Engineer Podcast

Episode: ⚡ Inside GitHub’s AI Revolution: Jared Palmer Reveals Agent HQ & The Future of Coding Agents
Guest: Jared Palmer (SVP at GitHub, VP Core AI at Microsoft)
Date: November 10, 2025

Episode Overview

This episode features a candid, wide-ranging interview with Jared Palmer, just 13 days into his dual leadership roles as SVP at GitHub and VP at Core AI, Microsoft. The discussion centers around the announcement of Agent HQ—GitHub’s ambitious new home for coding agents—exploring the evolution of AI-powered development tools, code generation models, and the emerging “Agent World.” The conversation also spans software workflows, model interoperability, developer pain points, and the importance of community-driven product iteration. Jared brings an insider’s perspective shaped by his journeys at Vercel, his work on V0, and his new mandate at GitHub.

Key Discussion Points

Palmer’s Background and Journey to GitHub

• Early Days at Vercel and AI Initiatives
  • Jared recaps building V0 (an agent focused on Next.js apps), contributing to the AI SDK, and fostering experimentation with frontend-focused tools.
  • On shifting from Vercel to GitHub:
    “At GitHub, we’re the home of all languages and frameworks and developers, so the scope is broadened… it’s just a different part of the map, if you will.” — Jared Palmer [01:46]
• Contrast Between V0 and GitHub
  • V0: Laser-focused on Next.js/React; rapid iteration possible within those constraints.
  • GitHub: Immense scale (>183 million developers), platform agnosticism, and broader strategic impact.

The Evolution of Coding Agents

• From Playground Prototypes to AI Agents
  • Jared describes how early projects like the “AI Playground” at Vercel laid foundations for experimenting with model APIs and UI paradigms.
    “What was cool about the AI Playground was it forced me to go through every single model provider’s API docs and figure out their quirks.” — Palmer [04:13]
  • Collaboration with community builders (e.g., shadcn) and the embrace of open-sourced templates expedited innovation.
• Rise of Tool Calls and the ‘Prompt to UI’ Paradigm
  • Initial hacks to approximate tool usage predated robust chat and context windows.
  • The “aha” behind V0: allowing AI agents to compose UI elements, documents, and code artifacts—foreshadowing today’s agent capabilities.

Model Choice, Agent Architecture, and Product Strategy

• Debating Composite Models vs. Selector UI [11:12–13:36]
  • Pros and cons of offering users a “model selector” versus building composite/synthetic models under branded names.
  • Independence from model labs as a business imperative; technical merit in leveraging strengths from different models.
  • Palmer describes GitHub’s vision:
    “At GitHub now, we are all about model choice… we also have Copilot, Copilot CLI, and third-party partners like Cloud Code, Codex, Cognition now like in Agent HQ. So you get the best of both worlds.” [13:11]
• Agent World vs. Chat World
  • The agent paradigm involves orchestrating APIs, file systems, for-loops, sandboxing, tool calls—far beyond simple chat UIs.
  • Strong coupling between model and agent logic is now essential; generic interfaces risk lowest-common-denominator performance.
  • On terminology: “I’m calling it Agent World… a loop with maybe compute runtime and like files.” — Palmer [14:06]

GitHub’s New Agent HQ: Vision and Early Moves

• What is Agent HQ?
  • Announced at GitHub Universe; envisioned as the "gravity well" for agents and developers—a collaborative, extensible platform.
• Seamlessness and Workflow Fluidity
  • Tight integration across GitHub, VS Code, Azure—aiming for a frictionless chain between tasks, PRs, code review, and automation.
    “One of the cooler things that AgentHQ can offer is this seamlessness, the spluidity with your workflow…fire off a task, and it creates a PR, but you can also open that PR up in VS Code in one click.” — Palmer [17:00]
  • AI “sprinkled” into native workflows—Salt Bae style—for everything from merge conflicts to CI/CD.
• Custom Agents and Extensibility
  • Announcement of custom agents, allowing tailored prompts and tools per workspace.
  • MCP (Model Capability Protocol) is framed as a key standard underpinning agent interoperability.

Developer Tooling, Standards, and Community Feedback

• Dev Containers, Sandboxing, and Setup Pain [18:29–21:50]
  • Discussion on the value and limitations of “dev containers” as a universal standard.
  • Widespread developer frustration with repo onboarding; Palmer notes, “You can’t predict what’s in the repo.”
  • Calls for open standards around auto-detection of frameworks and improved container onboarding.
• Standards Watch
  • Commentary on emerging protocols: Dev Containers, MCP, ACP, and their adoption barriers (especially the “PR problem”—marketing to developers).
  • Palmer’s take: MCP is widely adopted, especially with enterprise/AI-driven transformations.
• Model Reliability, Observability, and Quality Metrics [24:03–26:31]
  • The hardest challenge: pushing from 95% to “nines” of correctness/reliability.
  • Under-appreciated need for detailed observability: “We would do… every three hours rollup of key metrics and stats… like error-free sessions… especially now with agents which are multi-turn.” — Palmer [25:22]
  • Most teams operate “blind” to error rates and model infra reliability.
• Underexplored Areas: Data Analysis Agents
  • Surprise at lack of powerful data analyst agents (BI-style bots); coding agent skills increasingly leak into non-code tasks.

Memorable Quotes & Moments

• On Model Composability and Product Independence
  “One of the benefits of having your own branded model or synthetic or composite is that you can stitch these things together… you get to brand it. And you can decouple it from the launch of the Frontier Lab.” — Palmer [12:18]
• On Coding Agent Progress
  “When everybody else was trying to do general purpose coding agent, we were like no, we’re just going to focus on Next JS front end and Chad cn. And that allowed the team to focus.” — Palmer [10:17]
• On Future Agent & Workflow Vision
  “I think this is… this workflow where it’s just like seamless and fluid and you can stay in a flow state across… all devices, mobile web on GitHub.com, or in your local editor.” — Palmer [17:48]
• On Feedback and Community Iteration
  “All feedback is a gift, like it’s all a signal. And the more signal we can collect, the better decisions we can make and truly build this really, really useful website and company, like together.” — Palmer [35:15]

Specific Feature Topics

• Redesigned GitHub Homepage [29:04–30:18]
  • Launched to improve utility—responsive to a community critique that “the entire GitHub homepage is useless…”
  • Jared celebrates the internal team’s responsiveness, but notes, “More work to do, never done.”
• Stacked Diffs (Stacked PRs) [30:20–34:43]
  • Overview of why “stacked diffs” (a Facebook- and Graphite-inspired workflow) are highly requested: better for monorepos, large teams, complex code review.
  • History of several internal attempts at GitHub—blocked by complexity and perceived risk. Now “actively exploring” ways to support this in future roadmaps.

Timestamps for Key Segments

| Timestamp | Segment | |--------------|---------------------------------------------------------------| | 00:00–01:41 | Introduction; Palmer’s background/transitions | | 02:34–08:27 | The origin story of coding agents, Vercel, and V0 | | 08:27–10:21 | V0’s focus, early model limitations, and rapid growth | | 11:12–13:36 | Model choice, composites, business/product strategy | | 13:36–15:11 | Agent architecture: chat vs. agents, tool orchestration | | 16:24–18:29 | Agent HQ, seamless workflow, product vision | | 18:29–21:10 | Dev Containers, repo setup pain, standards discussion | | 21:36–24:03 | Container standards, CODING agent evolution | | 24:03–26:31 | Reliability, observability, measuring model effectiveness | | 26:31–28:00 | Data analyst agents, coding agents for non-code tasks | | 28:00–29:55 | Browser-based agents, pros/cons, developer workflows | | 29:04–30:18 | GitHub homepage redesign | | 30:20–34:43 | Stacked diffs: what/why, history at GitHub, next steps | | 34:43–35:40 | Philosophy on feedback, closing remarks, open DMs |

Notable Quotes (with Speakers and Timestamps)

• “At GitHub, we’re the home of all languages and frameworks and developers, so the scope is broadened… it’s just a different part of the map, if you will.”
  — Jared Palmer [01:46]

• “One of the benefits of having your own branded model or synthetic or composite is that you can stitch these things together… you get to brand it. And you can decouple it from the launch of the Frontier Lab.”
  — Palmer [12:18]

• “All feedback is a gift, like it’s all a signal. And the more signal we can collect, the better decisions we can make and truly build this really, really useful website and company, like together.”
  — Palmer [35:15]

• “I think this is… this workflow where it’s just like seamless and fluid and you can stay in a flow state across… all devices, mobile web on GitHub.com, or in your local editor.”
  — Palmer [17:48]

• “We would do… every three hours rollup of key metrics and stats… like error-free sessions… especially now with agents which are multi-turn.”
  — Palmer [25:22]

Conclusion: The Shape of AI Engineering to Come

Jared Palmer's energy, deep experience, and community orientation permeate this episode. The GitHub Agent HQ launch signals a new phase where agent-based coding, model composability, and seamless developer workflows are converging at an unprecedented scale. Technical, business, and cultural tradeoffs are all on the table—as are the feedback loops that drive user-driven innovation. The AI engineer’s toolbox is expanding, and listeners are invited not just to adopt, but to participate in shaping what comes next.

For full show notes and resources: latent.space