Spaii Labs, a small Barcelona-founded startup, opened public access to EasySpecs.ai on September 12. The pitch is exactly this blog’s thesis, arriving from a team with no reason to know or care that this blog has one: AI coding agents now produce code faster than any team can review it, so the quality gate has to move earlier — before code generation, not after. EasySpecs’ answer is a platform that turns your existing codebase into structured documentation, grounds every change request against that documentation before an agent touches it, and attaches machine-checkable and human-judged criteria to every resulting spec.
It’s a good idea, competently productized, entering a category that already has a bigger, free, open-source incumbent. Both things are worth taking seriously.
How it actually works#
EasySpecs connects to a GitHub, GitLab, Azure DevOps, or Bitbucket repository and analyzes it to produce what the company calls a “knowledge tree” — functional documentation of features, views, services, data models, infrastructure, and tests, with (per the company’s own claim) up to 98% line-of-code coverage assigned to some documented element. That’s the foundation: a machine-generated map of what a codebase actually does, kept separate from what anyone assumes it does.
From there, a change request gets “grounded” against that knowledge tree before any code gets written — the step Spaii Labs calls “polishing intent.” The resulting spec ships with two kinds of built-in verification, a distinction the company is clearly proud of and that’s genuinely useful framing: oracles, pass-or-fail conditions a machine can run automatically, and rubrics, judgment calls a human or an AI reviewer scores against explicitly stated criteria. Specs also carry explicit in-scope, out-of-scope, and — notably — unresolved sections, where ambiguous decisions are marked blocked rather than allowed to pass silently into implementation. That last detail is the part worth stealing regardless of which tool you use: an unresolved decision that fails loudly is a much better failure mode than one that gets quietly resolved by whatever the model guessed.
Once approved, specs flow outward into the rest of a team’s toolchain: into Jira and Linear as tracked work, into VS Code, Cursor, and Antigravity via IDE extensions, and into Claude Code, Codex, or GitHub Copilot as read-only context served over MCP. That MCP integration is the detail that places this squarely in the same ecosystem this blog covers constantly — EasySpecs isn’t trying to replace your coding agent, it’s trying to be the thing your coding agent reads before it starts working.
Pricing is tiered and modest for an individual repo: a free tier for connection and analysis, a €5/month-per-repository “Workbench” tier (launch pricing, normally €10), a €150/month “Factory” tier for organization-wide managed inference, and custom enterprise deployments for VPC or private cloud. No customer count or funding figures were disclosed at launch.
The competition it isn’t mentioning#
Here’s the part EasySpecs’ own launch materials don’t dwell on: it is not the first, the biggest, or the most-adopted tool solving this exact problem. OpenSpec — an open-source spec framework built by Fission-AI, launched via Y Combinator’s Launch YC program — has crossed 27,000 GitHub stars in under six months. Its approach is architecturally almost the opposite of EasySpecs’: instead of a hosted platform that analyzes your repo and serves specs back over MCP, OpenSpec adds a plain openspec/ folder directly into your repository — a spec library documenting capabilities in markdown, plus a “changes” system tracking proposals, designs, and implementation tasks as version-controlled files sitting right next to your code. It integrates with more than 20 AI tools, including Claude Code, Cursor, Windsurf, and Copilot, via slash commands rather than a managed backend, and it’s free.
That’s a real fork in how the spec-driven-development tooling category is settling: EasySpecs is betting that codebase analysis and spec generation are hard enough, and valuable enough as an ongoing service, to justify a paid SaaS layer with its own inference pipeline. OpenSpec is betting that specs are fundamentally just structured markdown that belongs in git alongside the code it describes, and that the framework should be free, portable, and owned entirely by the repository rather than a third-party platform. Both bets can be right for different teams. Neither has obviously won yet — 27,000 stars is real adoption signal that a Sept 12 launch simply hasn’t had time to generate, but stars aren’t the same as paying customers doing sustained SDD work, and EasySpecs’ repo-analysis depth (a genuine knowledge tree with coverage percentages, not just a markdown convention) is a capability OpenSpec doesn’t claim to offer.
Why this matters beyond one product launch#
This blog covered the broader trend back in April, when SDD went from niche methodology to a movement with 30-plus competing frameworks and AWS shipped a commercial SDD IDE in Kiro. EasySpecs is more evidence that the trend hasn’t peaked — it’s still producing new, differently-architected entrants five months later, and at least one of them (OpenSpec) has real, independently verifiable traction to show for it. The core argument this blog has made since its first article — that writing a spec an agent can reliably implement is the actual skill that matters now, more than writing the code by hand — doesn’t need EasySpecs or OpenSpec to succeed to be true. But it’s a healthier sign for the argument that two structurally different products are now competing to make that workflow easier, rather than one lone advocate insisting it should exist.
The honest caveat: EasySpecs is nine days old as of publication, pre-revenue-disclosure, and built by a two-person founding team (CEO Xesca Alabart, CTO Carlos Guirao Capistany) according to its own launch materials. Competent execution of a good idea is not the same as market validation. Worth revisiting once it has usage numbers to show, rather than a knowledge-tree feature list.
Sources: EasySpecs.ai launch coverage, TechBullion, Sept 2026 (direct fetch); EasySpecs.ai product site (primary, direct fetch); OpenSpec Launch YC page (primary, direct fetch); OpenSpec GitHub repository.
