AWS Veteran: The Agent Blueprint Behind 1,400 Engineers

"I need to stop using Opus. This doesn't work." That was Heitor Lessa's conclusion after a refactor cost him 200 million tokens, and it forced him to rebuild the entire agent workflow now used across 1,400 engineers. Heitor spent 11 years at AWS, built Lambda Powertools to 230 billion API calls a week, and in this episode he walks through the full SDLC workflow on screen, from discovery to merge check.

In this episode, we cover:

• The product loop: discovery, whiteboarding, and the /roadmap command
• Spec-driven development with Open Spec and why vanilla setups fail
• Three model tiers: SOTA for planning, mid-tier for implementation, cheap models for reviews
• Merge checks with adversarial reviewers and attestations that catch agents fabricating test results
• The /retro command: using the Socratic method to make your workflow more deterministic

If you're an engineer figuring out how to work with agents at team scale without losing trust in your codebase, this is the workflow to steal. This is also the first Beyond Coding episode with visuals on screen, so let me know what you think of the format.

Timestamps:
00:00:00 - The Math Doesn't Add Up
00:00:43 - Amazon Hypergrowth: 11 Years, 8 Different Roles
00:03:29 - Learning From the Trenches as a Technical Account Manager
00:08:38 - Developer Identity and the Birth of Lambda Powertools
00:10:20 - The Hard Parts of Working in Public
00:13:12 - How Powertools Hit 230 Billion API Calls a Week
00:16:42 - Career Advice: Learn Adjacent Roles, Not More Tech
00:19:37 - When Leadership Decisions Don't Make Sense to You
00:23:21 - The Product Loop Starts With Discovery
00:25:22 - From Whiteboard to /roadmap
00:27:37 - Why Humans Plan First and Agents Come Second
00:30:33 - Commands vs Skills Across 32 Different Models
00:33:38 - Adversarial Reviewers on Every Plan
00:36:07 - The Socratic Method, Explained
00:40:29 - Why He Only Takes Paper Notes
00:44:43 - The Five-Line Paper Trick for High-Stakes Meetings
00:48:18 - /new-work: Capturing Scope Creep Without Derailing
00:54:03 - The Dev Loop Begins: Open Spec Explore
00:56:34 - Three Model Tiers: SOTA, Mid, Cheap
00:57:43 - The $5,000/Month Per Engineer Question
00:58:57 - Guardrails vs Autonomy for 1,400 Engineers
01:04:22 - Auto-Sizer: Does This Task Even Need a Spec?
01:07:26 - Decision Fatigue and Why Frameworks Win
01:09:10 - The Plan Phase: Specs, Design, Formal Verification
01:13:07 - The Refactor That Cost 200 Million Tokens
01:15:11 - When Agents Forge Evidence They Ran Your Tests
01:17:27 - Local-First Architecture Explained
01:23:04 - The Apply Phase: Fully Autonomous Loops
01:24:30 - Coding Was Never the Bottleneck
01:26:39 - Why This Workflow Is an Investment
01:27:39 - Decision Logs and the /onboarding Command
01:29:06 - Running Agents Locally With Enterprise Governance
01:32:42 - Hooks: Making Quality Gates Deterministic
01:36:02 - Merge Checks: 15 Adversarial Reviewers Per Change
01:38:30 - /retro: Interviewing Yourself to Improve the Loop
01:43:12 - Trust, Loss of Trust, and Recovery With Agents
01:48:02 - Experience, Scars, and Critical Thinking
01:49:32 - Why Right Now Is the Time to Experiment
01:52:04 - Conviction Comes From Being in the Loop

#softwareengineering #aiagents #aws