What does it actually take to build a software factory around AI coding agents?

🔔Subscribe to AI That Works — a new episode drops every Friday. Follow the hosts too: Vaibhav (https://x.com/vaibcode)) and Dex (https://x.com/dexhorthy))

📍More from Boundary: they build BAML, the programming language for reliable, type-safe AI agents. Explore it at boundaryml.com, try it live at promptfiddle.com, or check out the code at github.com/boundaryml/baml.

In this episode, Dex and Vaibhav are joined by Tyler Brown and Cole Murray for a hands-on look at how real teams build AI-native software development systems — declarative planning, agent orchestration, development environments, verification, and the guardrails that let agents (and non-engineers) ship code.

👋 GUEST BIO & LINKS

About Tyler Brown: Co-Founder of BloomText, a HIPAA-compliant patient messaging platform for healthcare teams.
bloomtext.com | tbrown.io

About Cole Murray: Creator of Open-Inspect, an open-source system for running background AI coding agents in production.
github.com/ColeMurray/background-agents | murraycole.com

KEY TAKEAWAYS

• Better models require better harnesses.

• Human understanding is becoming the bottleneck.

• Development environments are incredibly difficult to automate.

• Non-engineers can ship code with the right guardrails.

• Software factories are ultimately about process engineering.

TIMESTAMPS

00:00 Teaser
00:22 Episode Overview
01:46 What Is a Software Factory?
04:00 How Companies Become AI Native
08:00 Tyler Brown Intro
09:16 Inside a Real AI-Native Software Company
12:08 The New Bottleneck: Planning, Orchestration, and Human Understanding
15:56 The Plan, Do, Review, and Improve Loop
17:49 How AI Agents Fit Into the Software Development Process
19:56 Why Verification Matters in AI-Native Development
21:21 Managing Multiple AI Agents From One Interface
25:57 How the Factory Turns Customer Bugs Into Code
27:54 Planning and Implementing a Fix With AI Agents
29:02 What Open-Inspect Is Trying to Solve
32:41 Why Development Environments Are So Hard
34:14 Building AI Environments Across Complex Codebases
36:41 Building the Full AI Agent Environment: Code, Context, and Production Data
38:49 Why Local Development Environments Don't Scale
40:29 Dev Environments and Building a Fast Testing and Iteration Loop
43:30 Tyler Brown's Hot Take: Sandboxes Are the Wrong Primitive
45:29 Letting Non-Technical Employees Use AI Agents
46:09 How Non-Engineers Earn More Autonomy With AI
48:52 Building Guardrails for AI-Native Development
50:16 Should AI Factories Outsource Human Learning?
54:05 The Future of AI-Native Software Development

RESOURCES / PEOPLE MENTIONED

• HumanLayer — Dex's company, "the multiplayer control plane for your software factory." humanlayer.dev

TOPICS COVERED

Software factories
AI coding agents
AI-native software development
Agent orchestration
Declarative planning
AI agent verification
Background agents
Cloud agents
Development environments
Multi-agent systems
Automated code review
Fast testing loops
Non-technical software development
AI adoption
Software engineering
Engineering process
AI-native organizations

HASHTAGS

#AIThatWorks #SoftwareFactories #AIAgents #CodingAgents #AIEngineering #SoftwareEngineering