CrystalSpec vs airfocus
Past the PRD, into the spec.
airfocus is a modular product platform — score with RICE or WSJF, roadmap the winners, and let AI draft a PRD. CrystalSpec picks up the moment that PRD stops being prose: a typed, versioned spec the team signs off on and coding agents read directly.
14-day trial, no credit card.
airfocus scores the PRD. CrystalSpec makes it a spec.
airfocus calls itself the first modular product-management platform, and the description fits. You assemble the parts you need: prioritization frameworks that score features with RICE, WSJF, or a formula you design yourself; Priority Poker for scoring together as a team; flexible roadmaps built from those priorities; a feedback and insights module; objectives and OKRs; and a documents capability where an AI Writer drafts a PRD — problem, objective, user story, requirements — straight into an item. Since Lucid Software acquired it in April 2025, it ships as airfocus by Lucid, and in mid-2026 it added a Product Intelligence Platform with an Insights Agent, a conversational agent, and a bidirectional MCP server. For deciding what to build and lining it up on a roadmap, it is a strong, configurable tool.
Here is the seam this page is about. A PRD in airfocus is prose inside a document. It can be scored, sequenced, and summarized — but it is still text. The moment engineering gets going, scope drifts, unforeseen edge cases turn up, and the document slips out of sync with what is actually being built. Nothing typed sits beneath it for a tool to resolve, there is no field-level trace of which decision moved between two versions, and there is no line drawn between the draft one person is editing and the version everyone else is trusting.
CrystalSpec's answer to the same feature is not a document at all. Here the spec is a lattice of typed entities: flows whose labelled branch points draw themselves into live, clickable diagrams; typed data-model fields that link out to other models; roles; coded test cases; and a glossary — all stitched into a shared terminology. The AI has a hand in authoring each of these, but strictly as proposals a person signs off on: it puts forward create, update, and delete suggestions already vetted for appliability, and you wave each one through or turn it away, the whole trail preserved. When you publish, the draft — held by one editor at a time — becomes a versioned revision that ships with an AI-written change summary and a field-level diff, reversible with its lineage intact. An inconsistency analyzer combs any slice of the project for clashes, omissions, and orphaned glossary terms, and a hosted MCP server plus a scoped GraphQL API open the whole graph to coding agents.
These are honestly different tools, not two versions of one. If your central problem is choosing and sequencing work across a portfolio, airfocus is designed for exactly that, and CrystalSpec does not try to compete on scoring or roadmaps — consult airfocus.com/pricing for its current plans. But if your central problem is that the PRD has to stay true for months while a team and its agents build against it, a prose document is the wrong container. Holding it together is the one job CrystalSpec exists to do.
A PRD you score — or a spec you build
airfocus can draft a PRD, score it with RICE, and drop it on a roadmap. CrystalSpec turns the same intent into typed entities your team approves and your agents can query.
airfocus — prose PRD, scored & roadmapped
CrystalSpec — typed, versioned spec
Modular PM platform vs spec workspace
| Dimension | CrystalSpec | airfocus |
|---|---|---|
| Category (mid-2026) | AI spec workspace | Modular product-management platform (airfocus by Lucid) |
| Prioritization frameworks | No: Not a feature | Yes: RICE, WSJF, custom scoring, Priority Poker |
| Roadmaps & portfolio | No: No roadmaps or portfolio views | Yes: Flexible roadmaps; portfolio (Enterprise) |
| Customer feedback / insights | No: Not a feature | Yes: Insights module + Insights Agent |
| Spec / PRD shape | Yes: Typed flows, models, roles, test cases, glossary | Partial: Prose PRD inside a doc/item |
| AI editing model | Yes: Appliability-checked proposals, approved row by row | Partial: AI Writer drafts into the doc directly |
| Versioning | Yes: Published revisions, field-level diffs, revert | Partial: Change history on items |
| Consistency checking | Yes: Inconsistency analyzer, findings graded by severity | No: None found as of mid-2026 |
| Flow diagrams | Yes: Generated live, branch-aware, clickable | Partial: Lucidchart diagramming (Lucid) |
| Agent access (MCP) | Yes: The typed spec graph, reached through hosted MCP, GraphQL, and webhooks | Yes: Bidirectional MCP to roadmaps, objectives, priorities |
| Tracker handoff | Yes: Atomic tasks dispatched into GitHub, Linear, ClickUp | Yes: Jira, Azure DevOps, Salesforce (Enterprise) |
| Pricing (mid-2026) | $10/seat/mo, 5,000 AI credits per member | Modular tiers; Professional + Enterprise, largely quote-based |
airfocus (by Lucid) features and pricing from airfocus.com and airfocus.com/pricing, as of mid-2026; airfocus was acquired by Lucid Software in April 2025 and its plans have since been restructured — verify current details. CrystalSpec figures from its live plan.
What a typed spec adds that a scored document can't
Typed entities, not headings
Branch-labelled flows, reference-aware data models bearing typed fields, roles, per-flow test cases, and a glossary — a linked terminology software can resolve on its own, rather than section headings only a reader can make sense of.
AI that proposes, never writes
airfocus's AI Writer types straight into the document; CrystalSpec's assistant is confined to proposals, each pre-screened for absent fields and dangling references. You wave one through or send it back, and the ones you send back stay logged next to whoever refused them.
A locked draft, a stable published version
One editor holds the draft while the rest of the team keeps working off the settled published copy. A publish files an AI-written summary and a field-by-field changeset, and any revision can be rolled back with its full lineage on the project's own timeline.
Contradictions surface before code does
Aim the inconsistency analyzer at an entire project, a single flow, or one step — with a goal attached if you like. Back come graded clashes, warnings, and glossary terms nobody uses, and “Fix all with AI” turns any of them into a reviewable proposal.
A typed graph agents query
The hosted MCP server, scoped GraphQL API, and HMAC-signed webhooks surface the structure itself: which steps a flow runs, the list of revisions, what changed between two versions, the fields on a model. airfocus's MCP hands over product data; CrystalSpec's hands over the spec.
From publish to tracked tasks
On publish, the AI breaks a revision into atomic tasks across GitHub, Linear, or ClickUp. Run the push again and no duplicate appears; each task keeps a pointer to the revision it came from — a one-way handoff with back-links, not a two-way sync.
What an agent actually gets back
Credit where it is due: airfocus met the agent era head-on. Its bidirectional MCP server, shipped in mid-2026 as part of the Product Intelligence Platform, gives assistants like Claude, ChatGPT, and Copilot secure access to structured airfocus data — roadmaps, objectives, and priorities — so an agent can reason about strategy without a human copying context into a prompt. For keeping AI aligned to what a product organization has decided to do, that is a genuinely useful bridge.
What sets the two apart is whatever the protocol hands back. Query airfocus and an agent gets product-planning data plus the prose of a PRD it still has to parse and interpret. Query CrystalSpec and the agent is reading typed structure: the steps that compose this flow and the points where it forks, which revision is currently published, what shifted at the field level between v6 and v9, the fields this model holds, the definition behind a glossary term. Every one of those answers is drawn from a single canonical published version, so ask the identical question from two agents and both come back with one and the same truth — never a pair of competing interpretations. Same standard, different substance.
Where each tool genuinely wins
Choose CrystalSpec if…
- What ships is an engineering-grade spec — typed flows, models, roles, and test cases — that stays under version control and stays queryable, not a prose PRD.
- You want the AI to draft while a logged human decision backs every change, paired with an analyzer that chases contradictions across the entire graph.
- Coding agents ought to pull typed entities and revision deltas over MCP and GraphQL rather than parse their way through a document.
- You want the publish step to do the filing — atomic tasks surfacing in GitHub, Linear, or ClickUp, each tied to the revision behind it.
Choose airfocus if…
- You need prioritization frameworks — RICE, WSJF, custom scoring, Priority Poker — to decide and sequence what to build. CrystalSpec has none of this, and airfocus is built for it.
- You want flexible, customizable roadmaps and portfolio views across many initiatives, plus a customer-feedback and insights module.
- Configurability matters more than a fixed structure: airfocus's modular platform bends to your process, framework by framework.
- Your delivery lives in Jira or Azure DevOps, and you value being inside the Lucid suite — Lucidchart diagramming, plus SSO and SCIM on Enterprise.
Questions teams actually ask
What is airfocus, and is it good at what it does?
airfocus is a modular product-management platform — now airfocus by Lucid after Lucid Software acquired it in April 2025. Its strengths are genuine: prioritization frameworks like RICE and WSJF, custom scoring, Priority Poker, flexible roadmaps, a feedback module, and AI that drafts PRDs. If ranking and roadmapping is the job, it is built for exactly that.
Can I bring my airfocus PRDs into CrystalSpec?
There is no import pipeline, but the path is simple. Drop the PRD's prose into CrystalSpec's assistant and it comes back with candidate flows, data models, roles, and test cases teased out of the text. You sign off on each one in turn, while the inconsistency analyzer surfaces the ambiguities the document had been quietly hiding.
Both products ship an MCP server now — what's the real difference?
airfocus's bidirectional MCP (as of mid-2026) exposes structured product data — roadmaps, objectives, priorities — to assistants like Claude, ChatGPT, and Copilot. CrystalSpec's hosted MCP and GraphQL API serve a typed spec graph instead: a flow's steps and decision branches, the revision list, the field-level delta between two versions, a model's fields, a glossary term. Same protocol, different payload.
Does CrystalSpec do prioritization frameworks or portfolio roadmaps?
No, and it is honest about that. CrystalSpec has no RICE or WSJF scoring, no Priority Poker, no portfolio roadmap, and no customer-feedback module. It starts after the decision to build, turning intent into a typed, versioned spec. For deciding what to build and sequencing a portfolio, airfocus is the purpose-built tool.
Where do the two AI editing approaches part ways?
airfocus's AI Writer drafts straight into an item's description — type a slash command and it produces a PRD with a problem, objective, user story, and requirements. CrystalSpec's AI is limited to submitting proposals; appliability runs first to catch absent fields and dangling references, and only then does a human approve or decline it, one entry after another. Every call is on the record.
Which is cheaper for a team, and how is airfocus priced?
airfocus has historically used modular per-editor tiers with free viewers; under Lucid its public plans have been restructured into Professional and Enterprise, largely quote-based as of mid-2026 — check airfocus.com/pricing. CrystalSpec sticks to a single plan billed at $10 a seat each month with 5,000 AI credits granted to every member, and looking over proposals costs no credits at all.
Does airfocus integrate with Jira, and does CrystalSpec?
airfocus integrates with Jira, Azure DevOps, and Salesforce — a real advantage if your delivery lives there. CrystalSpec does not touch Jira or Azure DevOps. What it does instead is turn a published revision's changes into atomic tasks dispatched to GitHub, Linear, or ClickUp, every one tied back to the revision it came from, with a repeated push producing no duplicate.
The bottom line
Choose airfocus to score, roadmap, and draft the PRD — it is a strong, configurable platform for exactly that. Choose CrystalSpec when that PRD has to become a typed, versioned spec your team approves and your coding agents can query.
14-day trial, no credit card.