The Pragmatic Engineer — selected conversations
The third golden age of software engineering – thanks to AI, with Grady Booch
Eighteen curated conversations on agent workflows, design, harnesses, and software quality
A manually reviewed selection of Pragmatic Engineer conversations with Maggie Appleton, Matt Pocock, Tibo Sottiaux, Martin Fowler, Dex Horthy, Boris Cherny, Mario Zechner and Armin Ronacher, Laura Tacho, Hillel Wayne, Mitchell Hashimoto, Kent Beck, Peter Steinberger, Grady Booch, Steve Yegge, DHH, Dax Raad, Charity Majors, and Addy Osmani. Topics include design prototyping, reusable skills, agent workflows, context and harness design, TDD and verification, architecture, measurement, accountability, and open-source governance. Practitioner observations, product-team claims, and forecasts are not controlled findings.
In this collection Browse 18 summaries 13 of 18
Gergely Orosz interviews Grady Booch about software engineering as systems trade-offs and AI coding agents as another rise in abstraction. Booch welcomes disposable personal software and agent-assisted learning, while arguing that durable systems still require fundamentals, architecture, and judgment across technical, economic, human, safety, and ethical forces 01:59–04:57(opens in a new tab), 53:50–56:46(opens in a new tab).
Key Points Covered
- Software engineering balances systems forces: Booch defines the work as finding reasonably optimal rather than perfect solutions across technical, economic, human, ethical, and other constraints; code is only one mechanism 01:59–04:57(opens in a new tab).
- Abstraction rises recur: He traces shifts from algorithms and data to objects, libraries, services, and platforms, with each layer enabling larger systems while making some prior skills less central 32:38–38:41(opens in a new tab), 46:46–48:44(opens in a new tab).
- Agents continue rather than begin the third age: In Booch's framing, AI assistance accelerates use of an already large ecosystem of packages and system components that emerged around the millennium 46:46–48:44(opens in a new tab).
- Disposable and durable software need different rigor: He celebrates single-user, throwaway automation, but conditions reduced concern about internals on the software not needing to endure 53:50–56:46(opens in a new tab).
- Agents reproduce recurring patterns well: Booch says familiar web and CRUD structures fit agent training better than the full breadth of computing; the episode offers examples rather than comparative evaluation 1:00:35–1:02:45(opens in a new tab).
- Guided generation can accelerate learning: He uses Cursor to create a minimal D3 example, studies it, and iterates, while stressing that the domain contains many precedents and that his own foundations help him judge the result 57:49–58:38(opens in a new tab), 1:03:41–1:05:33(opens in a new tab).
- Developers should move toward systems-level skills: Booch expects automation in delivery pipelines and simple apps, but sees continuing demand for people who manage complexity, architecture, scale, and human-technical interactions 1:06:26–1:09:05(opens in a new tab).
- Multi-agent coordination has precedents: He recommends studying Society of Mind, blackboards, global workspaces, and subsumption architectures rather than assuming agent orchestration is wholly new; these are directions, not evaluated implementations in the interview 1:09:05–1:13:10(opens in a new tab).
Full video: https://www.youtube.com/watch?v=OfMAtaocvJw(opens in a new tab)