BLOG
Thinking on AI-native engineering.
Research, analysis, and perspectives on coordination, governance, and the infrastructure of AI-native software teams.
Harness Engineering: The Discipline AI-Native Teams Need.
41% of AI agent deployments hit year-one ROI because of evaluation infrastructure. The test and governance harness is the layer most teams skip — and the one that determines whether their agents actually work.
Loop Engineering: You Design the System, Not the Prompt
Boris Cherny doesn't prompt Claude anymore. He writes loops. Loop engineering is the shift from manually prompting agents to designing automated systems that prompt, evaluate, and iterate on your behalf.
Test-Driven Development Is More Important Than Ever.
AI-generated code introduces 1.7-2.7x more defects. TDD and eval-driven development are the feedback signal that lets agents iterate without humans in every loop.
Agentic Engineering: The Discipline Behind AI-Native Teams.
57% of engineering orgs have agents in production. Two frontier models collaborating hit 25% success — half the single-agent rate. The gap is coordination, not capability.
The AI Engineering Coordination Layer: Worth More Than Your Coding Tools
AI coding tools make engineers 30-50% faster. Organizational delivery is flat. The gap is coordination waste, and the layer that fixes it saves more than the coding tools generate.
The Future of Software Engineering
Software engineering is shifting from human-coordinated workflows to AI-coordinated execution. This analysis maps the structural forces driving the transition — and the infrastructure it demands from every engineering organization.
From Human-Led to AI-Led: The SDLC Transition
The SDLC was designed for human execution agents — AI doesn't accelerate it, it restructures it entirely. Understanding the AI-native lifecycle is the defining engineering leadership challenge of 2026.
The AI Engineering Coordination Bottleneck
Adding more AI agents without solving coordination produces the same compounding chaos Brooks described for human teams. The bottleneck was never code generation — it's orchestration.
Why AI Coordination Complexity Grows Faster Than AI Velocity
As AI agent count scales linearly, coordination overhead scales combinatorially. Understanding why complexity outpaces velocity is essential to building systems that don't collapse under their own weight.
Why Orchestration Is the Next Software Infrastructure Layer
Every new execution model produces a coordination crisis, and every coordination crisis produces an infrastructure layer. AI-native engineering is following the same pattern that gave us Git, CI/CD, and Kubernetes.
How Stripe Built 1,300 PRs/Week — And What Everyone Else Can Learn
Stripe merges over 1,300 AI-written pull requests every week. The real differentiator isn't their AI model — it's a decade of infrastructure investment that most engineering teams haven't made yet.
AI Agents Aren't the Problem. Coordination Is.
Engineers blame AI agents for bad code, but the real issue is the environment. Without shared context, governance, and coordination infrastructure, even frontier models produce incoherent results.
Designing an Organizational Memory Graph for AI-Native Engineering
AI agents forget everything between sessions. An organizational memory graph gives them persistent, structured context — turning isolated executions into coherent engineering workflows across teams and time.
Adaptive Autonomy: Why 'Fully Autonomous AI' Is the Wrong Goal
The binary of full autonomy vs. full human control is a false choice. Adaptive autonomy — where AI agents earn trust incrementally through demonstrated competence — is the model that actually scales.
The Mythical Man-Month, Revisited for AI Agents
Brooks identified a fundamental truth about complex systems in 1975 — AI agents don't violate it, they amplify it. The same architectural solutions apply, now in modern multi-agent form.
Building the Design Partner Program: What We're Looking For and Why
We're selecting a small cohort of engineering teams to shape the coordination layer with us. Here's what makes a great design partner — and what both sides get out of the collaboration.
Multi-Agent Engineering: Patterns, Failure Modes, and What Coordination Actually Requires
A deep analysis of multi-agent system architectures in production engineering environments. We catalog the recurring failure patterns and derive the coordination primitives required to make multi-agent workflows reliable.
NEWSLETTER
Get LoomStack's engineering research in your inbox.
Bi-weekly. No marketing fluff.