Coding with Agents

The second season of Amp's documentary podcast follows the team's move from locally managed coding agents to remote-controllable, parallel agents running in isolated environments. The opening episode covers Amp Neo, ephemeral orbs, the declining importance of personalized local development, standardized team environments, and the implications for testing, review, CI, and software delivery.

Armin Ronacher and Ben Vinegar track the fast-moving practice and industry of agentic coding through monthly conversations about models, harnesses, software quality, open source, infrastructure, economics, and the changing role of developers. Episode 8 adds Pi creator Mario Zechner for a discussion of reinforcement learning, model regressions, open weights, trust, and the cost of keeping up.

Mario Zechner on building and using Pi

Mario Zechner demonstrates how he uses Pi to maintain Pi itself, pairing a deliberately minimal agent harness with strong human architectural control. The session covers minimalist tool design, self-modifying agent extensions, issue triage with GitHub CLI, deterministic checks, transcript-like branch workflows, and why human taste and module boundaries still matter when agents can generate code at high volume.

Thorsten Ball on Amp, shrinking harnesses, and coordinating many agents

Thorsten Ball argues that as frontier models improve, coding-agent products should remove clever harness scaffolding, expose models to simple tools and context, and focus on how humans organize many cheap, capable agents across new workflows.

Mario Zechner and Armin Ronacher on Pi, agent workflows, and engineering judgment

Mario Zechner and Armin Ronacher join Gergely Orosz to discuss Pi, a minimalist self-modifying coding agent; what 30 engineering-team interviews reveal about agent adoption; why automation bias, non-engineer code contributions, and agent-generated PRs stress review and open source; and why complexity, deliberate friction, and human judgment still matter in agentic workflows. The episode also compares MCP with CLIs and code execution as tool interfaces.

Maggie Appleton on team alignment and multiplayer agent environments

Maggie Appleton argues that coding agents make implementation cheap enough that team alignment becomes the bottleneck. The talk critiques single-player agent workflows and demos ACE, GitHub Next's prototype for shared sessions, cloud computers, previews, terminals, and agent collaboration.

Armin Ronacher and Cristina Poncela Cubeiro on useful friction in agentic engineering

Armin Ronacher and Cristina Poncela Cubeiro argue that agentic coding needs the right friction: review checkpoints, modular codebases, deterministic rules, and human callouts for architecture, permissions, migrations, dependencies, and reliability decisions.

Mario Zechner on Pi, self-modifying agents, and agent slop

Mario Zechner explains why he built Pi as a minimal, self-modifying coding-agent harness after frustrations with opaque context management and feature-heavy tooling. The talk covers Pi's small core and extension model, Terminal-Bench results, agent spam pressure on open-source maintainers, and a practical warning to slow down, scope agent tasks, and keep humans responsible for critical code.

Quinn Slack on Amp, Sourcegraph, and the business of frontier coding agents

Quinn Slack explains why Amp split from Sourcegraph's enterprise motion, why it targets frontier users instead of every developer seat, how model-lab subscriptions reshape coding-agent economics, and why AI pressure may disrupt traditional SaaS assumptions.

Mario Zechner on why Pi strips coding agents down to a malleable core

Mario Zechner explains why he built Pi after Claude Code, OpenCode, and other agent harnesses stopped matching his need for predictable context, observability, and extensibility. The talk covers agent history from ChatGPT copying to agentic search, Claude Code feature bloat, OpenCode context-management pitfalls, Terminal-Bench and Terminus as evidence for minimal harnesses, Pi's four-tool design and TypeScript extension model, and lightweight defenses against agent-generated open-source spam.

Mario Zechner on Pi, open agent harnesses, and developer control

Mario Zechner explains why Pi stays minimal, open, and extensible; how its TUI and extension model let users shape their own agent workflows; and why human architecture, review, and open ecosystems still matter when coding agents can generate code quickly.

Lucas Meijer on Pi, context management, and evaluating agent work

Lucas Meijer shares practical habits for agent-assisted software engineering, from asking agents for HTML reports and making repositories easier to navigate, to defining evaluation criteria before a run and requesting videos, screenshots, and slide decks that make review faster. The talk becomes a hands-on Pi demo focused on tree-shaped context management, throwing away unhelpful side quests, custom slash-command extensions, virtual-terminal visibility for long-running tools, and self-modifying software that can adapt to the user's current task.

Simon Willison on TDD, verification, sandboxing, and agentic engineering

Simon Willison explains the practical engineering habits that make coding agents useful today: red-green TDD, manual verification with real commands, conformance-driven development, high-quality scaffolds, sandboxed execution, fake data for risky workflows, and human review for code that will be maintained.

A multi-episode documentary series chronicling the Amp team's development of Amp, revealing groundbreaking insights on building practical coding agents. Topics include the shift from prompt to context engineering, the power of autonomous feedback loops, sub-agent architectures for context management, the "Oracle" pattern for multi-model reasoning, why codebases must adapt to agents, and evaluating LLM capabilities across providers. The team demonstrates why success depends on model quality over hidden tricks, treating agents as power tools rather than magic wands.

The rebranded Build Crew live series featuring live coding sessions, Prompt Duel at Dawn challenges, and opportunities to win Amp Credits. Continues the community-driven format with AI agent best practices and real-time development.

A series featuring discussions, sharing experiences, and exploring the latest trends in AI agents and development practices.

A series from the Amp team covering AI agent best practices and prompting strategies. Topics include structured prompts, chain of thought prompting, using the Oracle sub-agent for complex reasoning, external memory management with short threads, and practical tips for effective agent collaboration. Features real-world development scenarios and insights from the Build Crew community.

Beyang Liu in conversation with a16z

Beyang Liu discusses Sourcegraph's Amp agent and the shift in software engineering from coding to orchestrating agents, where 90% of time is spent on code review. Topics include the 'Smart' vs 'Fast' agent architecture, the rise of Chinese open source models for agentic tool use, and the transition to probabilistic software where developers manage intent while agents handle implementation logic.

Linus Torvalds shares his perspective on AI's impact on open source infrastructure, including crawler strain on kernel.org and AI-generated spam reports. He expresses skepticism about 'vibe coding' for serious maintenance while acknowledging AI as a great learning gateway for beginners. Compares AI to compilers—a tool that removes minutia rather than replacing developers—and predicts more software jobs, not fewer.

Quinn Slack & Thorsten Ball on Building Amp Code

Sourcegraph's CEO and Amp Dictator discuss the philosophy behind Amp Code, their unconventional 15x/day shipping cadence without code reviews, and critical insights on coding agent architecture. Topics include why subagents and prompt optimizers may be misguided, the diminishing importance of model selectors, agent-friendly tooling, version control in an AI-first world, and how agents are transforming enterprise software development workflows.

Beyang Liu at AI Engineer Conference

Beyang Liu explores the evolution of AI coding agents from co-pilots to autonomous tools, covering controversial design decisions (autonomous edits, minimal UI, model coupling), Amp's implementation, and power user insights. Discusses best practices like detailed prompts, feedback loops, and sub-agents for complex tasks, while highlighting anti-patterns including micromanaging and underprompting.