Panel Discussion: The Agentic Software Factory That Enterprises Actually Need
summary
Claim: giving engineers the ability to generate 10x more code does not yield 10x value and often shifts the bottleneck to trusting and verifying output. Panelists describe building a company context lake, a compiler-accurate view of repos, and isolated verification spaces, plus a review panel of agents encoding senior engineers’ tribal knowledge. They report non‑determinism compounding at scale (example: ~92% success vs 8% failure) and advise measuring what lands in production, not token consumption, with an ROI ratio of features/infra over incidents/bugs.
AI-written summary · about
chapters (21)
- 0:00Would 10x More Code Produce 10x More Value? Probably Not
- 1:05The Real Enterprise Question: Can You Trust AI-Generated Code?
- 2:38Panel Intros - Heena Purohit, Chase Norton, Itamar Friedman, and Patrick Vuong
- 5:13What Broke First as AI Code Scaled Across Organizations
- 6:44The Bottleneck Moved From Writing Code to Trusting Code
- 8:16What AI SDLC / Agentic SDLC Actually Means
- 9:18Judgment Spilled Downstream Onto Senior Developers
- 10:19Enterprise Reality: Thousands of Repos, Legacy Code, CI/CD, Compliance
- 11:52The 'Dark Factory' Vision and Working Backwards From It
- 14:27Code as Data: Why Agents Need a Compiler-Accurate View of the Codebase
- 15:59Context Is King - Company Context Lakes, Guardrails, Verification Spaces
- 18:32Bringing Determinism and Control Into a Probabilistic Stack
- 21:08Observe Before You Optimize, Then Govern
- 23:39Breaking the SDLC Into Small Components to Reduce Variance
- 25:13What Does Quality Even Mean? Breaking It Into Measurable Policies
- 27:19From Token Maxing to Outcome Maxing - What to Actually Measure
- 30:52Measure What Lands, Not What the Factory Consumes
- 32:26The ROI Equation: Features and Infra Over Incidents, Bugs and Cost
- 35:00Where Should Agents Be Autonomous - and Where Not?
- 38:04A Review Panel of Agents Encoding Senior Engineers' Tribal Knowledge
- 41:08Greenfield vs Brownfield and Building High-Confidence Verification Loops
related talks
01h 05mPanel: Agentic SE & Open Source in the FMware Era – Toward Trustworthy, Context-Aware FMware
50mPanel: Beyond The Vibe: Code Quality First
45mPanel Discussion: Ways to unlock engineering effectiveness
27mAgentic Systems: Leadership, Risk, and Accountability
20mDark Factories: Verification-First Agentic Engineering
46mPanel Discussion: Beyond vibe coding: Expanding AI’s role in delivery
33mThe Autonomous Enterprise
51mFuture of Software Development