Eleven Months In: What the AI Coding Landscape Actually Looks Like

Every consulting call starts with the same question: which AI coding tool should we adopt. Eleven months ago I would have guessed. Today I can answer with receipts. Eleven months of daily articles, thirteen production deployments, and four client migrations later, the landscape is no longer a fog. It has a shape.

The Terrain as I See It

The honest map has three zones. The first zone is interactive coding with a human in the loop: Cursor, Claude Code in terminal mode, Copilot inline. This zone is mature, boring, and profitable. The second zone is agentic execution where you direct an agent and walk away: Claude Code with subagents, Codex async tasks, background workers. This zone works but requires a skill most teams lack. The third zone is autonomous orchestration where agents spawn agents. This zone is hype. It works in demos and breaks in production.

Where the Money Is

Consulting billings tell the real story. Zone one generates most of my hours. Zone two generates most of my margin. Zone three generates most of my failed engagements.

• Zone 1 (interactive): 60 percent of hours, 40 percent of revenue
• Zone 2 (directed agents): 25 percent of hours, 45 percent of revenue
• Zone 3 (autonomous orchestration): 15 percent of hours, 15 percent of revenue, most of the rework

The economics say spend your learning time in zone two.

The Capability Ceiling

Every zone has a ceiling. Interactive coding ceilings at what a single skilled engineer can review per day. Directed agents ceiling at what a single skilled verifier can test per day. Autonomous orchestration ceilings at whatever your monitoring and rollback posture can absorb. If you know where your ceiling is, you know what problems to take on.

What I Got Wrong

In April I predicted that autonomous orchestration would be production-ready by year end. It is not. Reliability is still the bottleneck. Not capability. The models are smart enough. The harness around them is not disciplined enough. Anyone selling zone three today is selling futures.

What I Got Right

The skill soup thesis held. Generalists who can drive three zones beat specialists who master one. The builder mindset beat the engineer mindset. First principles beat best practices. Daily writing beat quarterly planning.

The Thesis for Year Two

If I were starting now, I would specialize in zone two: directing agents to do work I can verify. That is the commercially relevant sweet spot for the next twelve months. The Anthropic engineering posts keep confirming this.

Eleven months of writing produced a map. That is the real compound return.