While everyone argues whether Cursor beats Claude Code, they are missing that the whole conversation is stuck inside tier one. There are three tiers of AI coding adoption, and the leverage between them is not linear. It is a step function.
Tier one is AI-assisted coding. This is where the vast majority of developers live today. Cursor, Windsurf, Copilot — tools that help a human type code faster. They are excellent productivity multipliers, but they still require a human in the seat driving every keystroke. The AI is a faster typewriter, not a teammate.
Tier two is AI-guided development. Claude Code, Codex CLI, and the wave of agentic CLIs that followed. Here the AI takes more initiative — proposing architecture, writing larger blocks, making more decisions. But it still operates in interactive mode. The human is in the chair, watching the loop, approving each step. The AI is a sharp junior who needs constant supervision.
Tier three is autonomous background agents. Your site throws an error. Sentry catches it. An agent reads the stack trace, reproduces the bug, writes a fix, runs the tests, and opens a PR — all before any human notices. No keystroke required to start the work. The human shows up at review time, not at start time.
The gap between tiers two and three is where Tembo is focused, and it is the only gap that actually matters. Tier-one tools win the productivity argument. Tier-three agents win the leverage argument — and leverage is what compounds. The companies still benchmarking tier-one autocomplete against itself in 2026 are going to wake up to teams that have been quietly running tier three for two years.
If you want to know where this is going, stop asking which tier-one tool is best. Start asking what your team would build if half the work fired itself off events instead of off humans. I have argued elsewhere that the road ahead is platforms, not tools — the tier-three opportunity is the platform position, and almost nobody is building for it yet.
— Ry
Sources
Related Essays
The Road Ahead for AI Coding
We are still early. The trajectory is clear. The companies that build platforms — not just use tools — define the next decade of software development.
Democratizing Software Engineering
AI coding is not about replacing developers. It is about expanding who can request software work in the first place — and 10x-ing total demand.
From Code Writer to Code Judge
The biggest blocker to AI coding adoption is not the technology. It is the developer's ego attachment to typing. The role has to shift from author to judge.
Key takeaways
- Tier one is autocomplete with attitude. Tier three is autonomous repair while you sleep.
- Most developers are still operating in tier one without realizing tier three is already shipping in the wild.
- The leverage compounds non-linearly across tiers — each one removes a different category of human bottleneck.
FAQ
What separates tier two from tier three?
Tier two still requires a human in the chair driving the loop. Tier three fires on events — a Sentry error, a Linear ticket, a failing test — with no human present until review time. The shift from interactive to event-driven is the real discontinuity.
Where are most companies today?
Almost everyone is in tier one — Cursor, Copilot, Windsurf used as faster autocomplete. A small slice has crossed into tier two with Claude Code and Codex CLI. Tier three is barely populated, and that is the entire opportunity.