The Post-Prompt Era: Alex Graveley on Delegating Work to AI
In this episode of the ODSC Ai X Podcast, Sheamus McGovern speaks with Alex Graveley, co-creator of GitHub Copilot, former contributor to Perplexity Computer and Comet, and now founder of UFO AI. Alex has helped build several of the products that shaped how people work with AI, from AI-assisted coding to agentic systems that can use computers, maintain context, and take action across workflows. He shares how his view of AI has changed since the early days of GitHub Copilot, when he worked closely with OpenAI before ChatGPT and was still skeptical of the bigger claims around AGI. Listen to this episode to understand what “delegating work to AI” actually means beyond the buzzword. Alex breaks down the practical pieces that make agentic systems useful in real organizations: shared memory, parallel agents, trust thresholds, sandboxing, model routing, and the shift from prompting every step to supervising AI systems that can carry more of the workload. The conversation looks at what comes next as AI shifts from copilots and chat interfaces toward systems we can delegate real work to, and why AI-first companies may be much smaller, faster, and more software-defined than the companies that came before them. 0:00 GitHub Copilot, Perplexity, UFO AI, and the future of AI agents 1:26 Alex Graveley on building GitHub Copilot, Comet, and UFO AI 3:15 From AI copilots to autonomous AI agents 5:28 Why AI agent harnesses matter as models improve 6:04 Shared memory, enterprise context, and AI agent infrastructure 7:23 Lessons from Perplexity Computer and Comet 8:00 UFO AI, startup workflows, and organizational memory 9:07 Running 20 AI agents in parallel 10:24 How to manage multiple AI agents across Slack, web, and terminal 11:33 Instinct, personal AI assistants, and long-running tasks 12:17 Self-recovery and Anthropic’s agent advantage 13:00 How much work can we delegate to AI models? 14:28 Context, trust, and letting AI agents take action 15:49 High-trust vs. high-control AI workflows 16:48 How AI is changing software engineering 17:08 Why most AI engineers write less code now 18:37 Prompting vs. true AI delegation 19:36 How Perplexity and UFO write code through Slack 20:24 AI memory, company context, and institutional knowledge 21:44 Trust, control, and sandboxing AI agents 22:22 Why sandboxes are table stakes for AI agent security 24:10 Docker, Firecracker, and cloud sandboxes for AI agents 25:22 The future of software and custom AI-built apps 26:34 Lovable, vibe coding, and AI-maintained applications 27:34 AI supervisors, custom software, and smaller companies 28:25 How AI agents could power micro-companies 29:13 Personal AI, voice interfaces, Instinct, Muse, and OpenClaw 30:30 The future of personal AI vs. workplace AI 32:28 Open-source models vs. frontier models in AI agent systems 33:22 Model routing, OpenRouter, and choosing the right model 34:43 Startups, niches, and the next wave of AI products 35:10 What to expect from UFO AI 36:16 Self-hosting, business memory, and enterprise AI deployment 36:46 Closing thoughts on UFO AI and the future of agentic work Visit our website and choose the nearest ODSC AI event to attend and experience all our training and workshops: https://odsc.ai To watch more videos like this, visit https://aiplus.training Sign up for the newsletter to stay up to date with the latest trends in data science: https://opendatascience.com/newsletter/ Follow us online! • Facebook: https://www.facebook.com/OPENDATASCI • Instagram: https://www.instagram.com/odsc/ • Blog: https://opendatascience.com/ • LinkedIn: https://www.linkedin.com/company/open-data-science/ • X (twitter): https://x.com/_odsc