The comparison pages on this site cover thirteen AI agents, coding tools and app builders, each against TODO for AI and each against the rest of the field. They are generated from a single data file, so every page uses the same facts, the same six dimensions, and the same ranking rule. This guide explains that rule and indexes the pages.
Disclosure: TODO for AI builds these pages about its own competitors. Every fact carries a verification date and the scoring weights are published below so the ranking can be checked rather than trusted.
The tools covered
| Tool | Category | Models | From | Compare |
|---|---|---|---|---|
| Claude Code | coding agent | Anthropic only | $20/mo | vs · alternatives |
| OpenAI Codex | coding agent | OpenAI only | $8/mo | vs · alternatives |
| Cursor | IDE | multi-vendor, metered | $20/mo | vs · alternatives |
| OpenCode | coding agent | any provider (BYOK) | free | vs · alternatives |
| Cline | IDE extension | any provider (BYOK) | free | vs · alternatives |
| Aider | coding agent | any provider (BYOK) | free | vs · alternatives |
| OpenHands | coding agent | any provider | free (self-host) | vs · alternatives |
| Claude Cowork | general agent | Anthropic only | $20/mo | vs · alternatives |
| ChatGPT | general agent | OpenAI only | free | vs · alternatives |
| Lindy | general agent | most major models | $30/mo | vs · alternatives |
| Manus | general agent | vendor-selected | $20/mo | vs · alternatives |
| Lovable | app builder | vendor-selected | $25/mo | vs · alternatives |
| Base44 | app builder | vendor-selected, choice on paid plans | $16/mo | vs · alternatives |
Prices are public list prices at the verification date shown on each page (15–18 September 2026). “BYOK” tools are free to run but bill tokens through your own provider account.
How a page is scored
Each tool receives an integer score from 0 to 5 on six dimensions. The same six apply to TODO for AI.
| Dimension | What it measures |
|---|---|
| Model freedom | Which vendors and models you can run |
| Long-running work | Persistent machine, scheduling, parallel tasks |
| Beyond code | Marketing, ops, browser, email — not only repositories |
| Openness | Source available, self-host, extend |
| Team & permissions | Roles, pooled usage, tool approvals |
| Cost predictability | Flat plan versus metered tokens |
A direct comparison (/vs/<tool>) shows both tools’ scores side by side with a feature matrix and the pricing ladder. It does not rank; it presents the two profiles and states who should choose which.
How a role guide ranks
The six audience guides — founders, solo developers, agencies, marketing teams, non-technical teams, startups — are ranked, and the ranking is computed rather than written.
- edits code?
- acts in business systems?
- usable without a terminal?
- each dimension gets w ∈ [0.25, 3]
- unlisted dimensions weigh 1
- tools sorted descending
where is the six dimensions above and defaults to 1 for any dimension the role does not list.
The gate runs first, for roles that declare one, because a weighted sum lets a strength compensate for a task the tool cannot perform at all. The marketing-teams guide requires both acts in business systems and usable without a terminal; a tool that fails either is listed as excluded rather than ranked low, however well it edits code. The founders and startups guides declare no gate and rank the full field.
Two example weight vectors from the published data:
| Role | models | autonomy | scope | openness | team | cost |
|---|---|---|---|---|---|---|
| Founders | 1.5 | 2.5 | 3 | 1 | 1 | 2 |
| Solo developers | 3 | 2 | 0.5 | 2 | 0.25 | 1.5 |
The same tool lands in different positions on different guides because the weights differ, not because the facts do. Each guide page states its weights in prose so the reader can disagree with them.
Reading the pages
- Start from the tool you already use: the vs page answers “what changes if I switch”; the alternatives page answers “what else is there”.
- Start from the job: the role guide answers “which tool first”, with the weighting made explicit.
- For an editor evaluated on its own terms rather than against us, read the Cursor review.
- For the benchmark number that appears on the coding-agent pages, read the Terminal-Bench 2.1 methodology — including why it is not directly comparable with the leaderboard values shown beside it.
One-page versions
Five of the comparisons condensed to a single table each, as public Google Docs: AI to-do list: the 5-item test, Claude Cowork alternatives for teams, back office: who finishes the job, agent in your logged-in browser, Claude Code for non-coding tasks. Index sheet.