Rankings of AI coding tools are worthless by the time they are indexed. The models change monthly, the products ship weekly, and the benchmark everybody quotes was not measuring your work.
Short answer: this site publishes durable guides about working with AI coding agents, deliberately avoiding the things that expire. No numbered rankings, no benchmark scores quoted as verdicts, no pricing figures, no version numbers. We compare tools by shape, explain the mechanics that stay true across releases, and are honest that the gap between the leading options is smaller than the gap between using one well and using one badly.
What we think is actually true
Three positions the whole site rests on.
The loop is the thing. Every agent reads context, plans, calls a tool, observes the result and revises. Everything vendors compete on sits inside one of those five steps. Understanding that makes the market legible, and it survives every model release.
Most capability complaints are context complaints. An agent cannot see what you did not give it, does not know your conventions unless they are written down, and does not remember last session. A large share of “the agent is bad at this” is really “the agent was given nothing to work from”. The fixes are unglamorous and they work.
Verification is now the bottleneck. Writing code used to be the slow part. That ratio has inverted. How much agent work a team can safely absorb is entirely a function of how much of the checking is automated, which is why we write more about tests and review than about prompts.
What we publish
Twelve guides across six threads:
- Fundamentals. What an agent is, how the loop works, and why context management decides output quality.
- Comparisons. Tools sorted by shape and by what they were built for, including where each is the wrong answer.
- Workflow. Instruction files, and the terminal-versus-editor split that decides how you work.
- Code Quality. Reviewing code whose author cannot explain it, and verifying work you did not write.
- Operations. Permissions, secrets, sandboxing, blast radius, and where the tokens go.
- Integrations. MCP, tool interfaces, and the infrastructure an agent quietly commits you to.
Independence
We take no payment for coverage, placement or links. We run no affiliate links. We have no commercial relationship with any tool covered here, and nothing on this site is ranked because of one.
That is worth stating plainly because this category is full of content that exists to sell something. Our comparison posts are written to be useful to somebody who ends up choosing the tool we did not lead with, and they usually conclude that either option is fine and the repository matters more.
Where a post recommends something, it also states the conditions under which the recommendation does not hold. The full method is on editorial standards.
What you will not find here
- Numbered rankings. “Top 10 agents” is a format that guarantees being wrong within a quarter.
- Benchmark scores as verdicts. Public benchmarks measure something real and specific, usually not your workload, and they get optimised against once they become marketing assets.
- Pricing figures. They change, and a stale price is worse than a link to the current one.
- Version numbers in post bodies. Same reason.
- Prompt libraries. A prompt that works this month is not knowledge.
Who writes here
Six people, each with a standing beat rather than a byline invented per article:
- Nils Aaberg, agent fundamentals, autonomy models, how agents plan
- Anjali Sethi, tool comparisons, editor and terminal ergonomics
- Marcus Delaney, context, memory, instruction files, repo-level prompting
- Yuki Shimada, code quality, review, testing, verification
- Ingrid Bauer, security, permissions, secrets, sandboxing, cost control
- Tomas Ferreira, protocols, integrations, infrastructure choices
Bios and every post by each of them are on the authors page.
Where to start
What an AI coding agent actually is is the pillar. If you are choosing a tool, the best AI coding agents sorted by what they are for. If you already have one and the output is disappointing, skip the comparisons entirely and read writing an instruction file your agent will actually follow.
Corrections and topic requests are welcome, and the process is on the contact page.