Uploaded July 2026 | Updated September 2026, 11 hours ago
Everyone’s agreed that AI writing code is only about 20% of shipping software. Chris McClanahan runs the other 80%. He’s a product lead over Tanzu Platform, and about four months ago his org started pointing the same AI tools they tell customers to use back at their own product management. Chris and @thecote walk through what actually happened: the “Centaur pods,” the “intent director” agent that replaced the PM’s Jira paperwork, and the roughly three weeks where the whole thing was, in Chris’s words, a clogged pipe with a trickle of water coming out.
It started as a disaster, as Chris says. The first run asked for an airplane and got a wheelbarrow, because the agent knew nothing about the platform’s ten-year history. What fixed it wasn’t a smarter model. It was onboarding the AI like a new hire: feeding it the platform’s history in order of authority, then giving it boundary conditions instead of hoping it would read your mind.
Chris McClanahan is on LinkedIn: linkedin.com/in/chris-mcclanahan-579a081
Interested in a great platform as a service for your enterprise apps? Try Tanzu: vmware.com/products/app-platform/tanzu
Topics covered:
- Why “programming is only 20% of the work” is the interesting part, not the objection
- Centaur pods and the intent director: what a product manager’s job becomes when the forms go away
- BMAD (Build More Architecture Dreams) as the open-source pipeline they married to their existing release process
- Going from six-month release cycles to a 24-hour throughput target (36–48 hours on an average day)
- The airplane-and-wheelbarrow problem: an AI is only as good as the context in front of it
- Onboarding the AI: docs, knowledge base, customer usage, then blogs - in that order of authority
- Moving people’s cheese: the org fight when four PMs collapse into one pod, and the hurt feelings that come with it
- Why the useful “patch” is now a markdown reminder of what not to do, not a code change
- The AI telling you “nobody’s gonna use this” - and why only about 10% of your great ideas survive contact with the usage data
- AI fatigue, staring-AI syndrome, and Purnima Padmanabhan on institutionalizing the “challenger” role so humans don’t rubber-stamp beautiful-but-wrong output
Index:
00:00 The other 80% of shipping software
01:20 Meet Chris McClanahan
02:10 What Tanzu Platform actually is
04:43 Centaur pods, and why they started as a disaster
09:05 What’s in a pod: the intent director
12:55 BMAD: Build More Architecture Dreams
13:58 Chasing a 24-hour throughput
15:14 Tracing a feature through the pipeline
20:53 Killing the Jira paperwork
22:34 The airplane and the wheelbarrow
24:33 Onboarding the AI like a new hire
29:09 Moving people’s cheese
36:15 Prototypes don’t have to be pretty
41:37 Ask it to ask you questions
45:32 AI fatigue and the challenger role
51:51 Six weeks is a year and a half
Everyone’s agreed that AI writing code is only about 20% of shipping software. Chris McClanahan runs the other 80%. He’s a product lead over Tanzu Platform, and about four months ago his org started pointing the same AI tools they tell customers to use back at their own product management. Chris and @thecote walk through what actually happened: the “Centaur pods,” the “intent director” agent that replaced the PM’s Jira paperwork, and the roughly three weeks where the whole thing was, in Chris’s words, a clogged pipe with a trickle of water coming out.
It started as a disaster, as Chris says. The first run asked for an airplane and got a wheelbarrow, because the agent knew nothing about the platform’s ten-year history. What fixed it wasn’t a smarter model. It was onboarding the AI like a new hire: feeding it the platform’s history in order of authority, then giving it boundary conditions instead of hoping it would read your mind.
Chris McClanahan is on LinkedIn: linkedin.com/in/chris-mcclanahan-579a081
Interested in a great platform as a service for your enterprise apps? Try Tanzu: vmware.com/products/app-platform/tanzu
Topics covered:
- Why “programming is only 20% of the work” is the interesting part, not the objection
- Centaur pods and the intent director: what a product manager’s job becomes when the forms go away
- BMAD (Build More Architecture Dreams) as the open-source pipeline they married to their existing release process
- Going from six-month release cycles to a 24-hour throughput target (36–48 hours on an average day)
- The airplane-and-wheelbarrow problem: an AI is only as good as the context in front of it
- Onboarding the AI: docs, knowledge base, customer usage, then blogs - in that order of authority
- Moving people’s cheese: the org fight when four PMs collapse into one pod, and the hurt feelings that come with it
- Why the useful “patch” is now a markdown reminder of what not to do, not a code change
- The AI telling you “nobody’s gonna use this” - and why only about 10% of your great ideas survive contact with the usage data
- AI fatigue, staring-AI syndrome, and Purnima Padmanabhan on institutionalizing the “challenger” role so humans don’t rubber-stamp beautiful-but-wrong output
Index:
00:00 The other 80% of shipping software
01:20 Meet Chris McClanahan
02:10 What Tanzu Platform actually is
04:43 Centaur pods, and why they started as a disaster
09:05 What’s in a pod: the intent director
12:55 BMAD: Build More Architecture Dreams
13:58 Chasing a 24-hour throughput
15:14 Tracing a feature through the pipeline
20:53 Killing the Jira paperwork
22:34 The airplane and the wheelbarrow
24:33 Onboarding the AI like a new hire
29:09 Moving people’s cheese
36:15 Prototypes don’t have to be pretty
41:37 Ask it to ask you questions
45:32 AI fatigue and the challenger role
51:51 Six weeks is a year and a half
