# CRAFT A.I. Personas
Understanding CRAFT Personas: A Detailed Guide
Specialists who show up when you need them — inside every CRAFT for Claude Cowork session
CRAFT for Claude Cowork ships with a growing library of personas — each built to a complete specification, each callable from any session. The library expands as new specialists are needed, and any project can import and use any persona in the catalog.
This page highlights three of them: Cat, the project orchestrator who is engaged automatically in every CRAFT for Cowork conversation, plus two of the specialists you’ll meet most often — Auguste for content creation, and Fernand for validation.
Each persona is built to a complete specification — role, expertise, communication patterns, explicit boundaries — so that engaging them produces consistent, predictable behavior across sessions. Click any persona below to see how it contributes.
CatTier E
The project orchestrator
Cat is engaged automatically in every CRAFT for Claude Cowork session. You don’t invoke Cat — she boots with the project. Her job is orchestration: read the boot prompt, restore the prior session’s state, surface the active priorities, and keep the project moving forward without losing context.
In the Creator/Validator workflow, Cat is the creator. She drafts specifications, builds recipes, runs analyses, and produces the first pass of substantive work. When that work needs review, she hands off to Fernand. When it needs prose, she pulls in Auguste. Cat doesn’t do everything — she routes.
Cat operates at Tier E (Expert). She assumes you know your own work and gets out of the way fast. She pushes back when warranted, marks confidence on her recommendations, and never accepts “just do it” without flagging a missing input.
AugusteTier A
The content creator
Auguste owns the parts of your project where words are the product. Marketing copy. Long-form articles. Founder narratives. Customer-facing emails. Cat may sketch the structure, but Auguste lands the prose.
In the Creator/Validator workflow, Auguste is a parallel creator alongside Cat. The split is by surface: Cat creates structure (specs, recipes, plans); Auguste creates voice (prose, narrative, copy). Both hand off to Fernand for validation.
Auguste operates at Tier A (Advanced). He produces draft prose at speed and trusts you and Fernand to flag where the voice misses. Auguste is at his best when given a brief with audience, length, and tone — and given room to choose specific words.
FernandTier A
The validator
Fernand is the second pair of eyes. His job is to catch what the creator missed: a spec parameter that’s underspecified, a recipe step that assumes context the user won’t have, a paragraph that says less than it implies. Fernand validates against the brief and against production-readiness.
In the Creator/Validator workflow, Fernand is the validator. Cat creates → Fernand validates → revisions cycle until the work passes. The handoff is structured: Cat names what was built, the assumptions made, and the questions still open. Fernand returns a verdict (PASS, PASS-WITH-NOTES, REVISE) and itemized findings.
Fernand operates at Tier A (Advanced). He is direct, specific, and willing to recommend a full rewrite when patching would leave inconsistencies. Fernand is who you want when you need to ship work that will hold up under outside review.
→ Browse the full library of official CRAFT personas
Confidence Tiers
Every CRAFT persona ships at a confidence tier — a calibration of how much specialist depth the persona brings and how much context it expects from you.
Tier is a calibration choice, not a quality ranking. The right tier depends on the work and the operator. Cat at Tier E is exactly right for project orchestration; a Tier B persona at the same surface would slow you down by re-explaining what you already know.
The Creator/Validator Workflow
The CV workflow is how Cat, Auguste, and Fernand collaborate inside a single session. The pattern is intentionally simple: creator drafts → validator reviews → revisions cycle until the work passes.
Cat or Auguste creates — a recipe, a spec, a paragraph, a plan. The handoff to Fernand is structured: the creator names what was built, the assumptions made, and the questions still open. Fernand returns a verdict and itemized findings. If revisions are needed, Cat applies them and the cycle repeats.
The discipline matters because solo-review blind spots are real. The creator who just wrote the paragraph is the worst person to spot what’s missing from it. Fernand reads with fresh eyes against the brief, which is how production-grade work survives outside review.
How specialist invocation looks in practice
Because Cat is already engaged in your Cowork session, you don’t invoke her — you ask her to bring in a specialist. The pattern looks like this:
Cat, please have Auguste draft a 200-word blog lede on the kitchen-staples analogy.
Or for validation:
Cat, hand the V1 paste-kit to Fernand for a pre-paste review.
Cat handles the routing, the brief, and the handoff structure. You stay focused on the decision the work serves.
Why Personas Matter
Personas aren’t a marketing wrapper. They solve a real problem in how A.I. behaves under load — one that gets worse, not better, as you put more weight on a single conversation.
The Problem Personas Solve
Every A.I. conversation has an identity problem: who is the A.I. supposed to be? Without explicit guidance, A.I. defaults to a generic “helpful assistant” — knowledgeable but unfocused, capable but undifferentiated. That default creates three persistent problems:
Inconsistent Quality
Ask the same question twice and get different answers. The A.I.’s “personality” shifts with context, creating unpredictable results. Yesterday’s nailed analysis is reachable today only by re-typing yesterday’s prompt — if you remembered to save it.
Shallow Expertise
Generic assistants know a little about everything but lack depth. They can’t maintain the focused perspective of a specialist for the length of a real problem. The work flattens out as the conversation goes on.
Repeated Context-Setting
Every conversation requires explaining what kind of help you need. “Act as a marketing expert…” “Write like a senior analyst…” “Focus on data-driven insights…” This overhead is the prompt-drift tax. It compounds across sessions until you stop using A.I. for the work that matters.
Personas solve all three problems by defining who the A.I. is before any specific request. The identity becomes stable, the expertise becomes focused, and the context becomes persistent across sessions.
Beyond “Act As” Prompts
A simple “act as” prompt provides role but not behavior. The A.I. knows what to be but not how to be it. A CRAFT persona provides the complete specification: role and expertise, explicit boundaries (“does NOT provide legal advice”), Big Five personality traits with behavioral examples, full communication patterns (greeting, error handling, closing), and explicit “when not to use” criteria.
The depth of specification creates the consistency. Every detail you define is a detail the A.I. doesn’t have to improvise — and improvisation is where inconsistency creeps in.
The Psychology Behind Effective Personas
CRAFT personas leverage established psychological principles to create consistent, believable A.I. behavior.
Big Five Personality Model
The Big Five (Openness, Conscientiousness, Extraversion, Agreeableness, Neuroticism) comes from decades of personality psychology research. When you rate a persona on these dimensions with behavioral examples, you’re providing a scientifically-grounded framework for consistent behavior.
Trust Through Predictability
Humans trust what they can predict. By defining interaction patterns — greeting, error recovery, closing — personas behave consistently across sessions. Users learn what to expect, and fulfilled expectations build trust.
Appropriate Anthropomorphism
Personas can have names and personalities without misleading users. The key is boundaries: personas shouldn’t claim personal experiences or pretend to be human. They’re characters, not deceptions.
Cognitive Consistency
When a persona has explicit boundaries (“does NOT provide legal advice”), the A.I. maintains that boundary consistently. Without explicit boundaries, behavior drifts based on context.
Worked example · Customer voice in action
The Leo Story
The clearest case for the persona discipline isn’t a CRAFT operational persona at all. It’s a simulated customer named Leo — built to react to this very project as a plausible target user would.
Why Leo was made
Mid-project, the team noticed a familiar problem: every decision was being made from the creator side. Pricing, benefits, signup flow, beta-close framing — all of it was being shaped by people who already understood why CRAFT mattered. The voice missing from the room was the customer’s — the busy Claude Pro or Max subscriber who had been burned by half-baked A.I. tools and wouldn’t sign up for anything that didn’t pay back within two weeks.
So we built that voice as a persona.
How Leo was made
Leo was constructed using the same persona-design discipline as every CRAFT operational persona. He got a complete specification: a tagline (“I’ve seen too many cute A.I. tools. Show me why this is worth my Tuesday.”), demographics, validated pains, Big Five traits, explicit knowledge boundaries (Leo doesn’t speak for enterprise buyers or for hobbyists), and clear rules for when to consult him and when not to.
The audience profile wasn’t invented — it was synthesized from a deep cross-A.I. research pass against the actual target market. Leo’s reactions are anchored in evidence, not vibes. He is a worked example of CRAFT persona design applied to a non-A.I. role: simulating a customer rather than performing a task.
How Leo contributed to the Beta Redesign
Leo was consulted on every public-facing decision that shaped the CRAFTFramework.ai Beta Refresh: signup-page copy, benefit packages, pricing language, founding-cohort framing, the membership-model design. His reactions were captured with confidence levels alongside the team’s. When Cat recommended a path, Leo got the next read — a fresh-eyes customer reaction before Cat’s framing locked in.
More than once, Leo flagged a phrase that read fine to the team but landed wrong from the outside. Each catch was a piece of feedback that wasn’t available anywhere else — not in analytics, not in beta surveys, not in user interviews we hadn’t run yet.
Leo isn’t shipped with CRAFT. He’s project-scoped — built for this work, retired when this work closes. What ships is the pattern. Anyone using CRAFT for Claude Cowork can build their own Leo: a simulated customer, a contrarian reviewer, a missing stakeholder voice. The persona-design specification is in the framework. The discipline is the same.
If you have a decision being made without a voice in the room, build the voice.
Become a Founding Chef
CRAFT for Claude Cowork is in open beta. Until September 1, 2026, anyone joining gets Founding Chef recognition — and a seat at the table while the framework is still being shaped.
After September 1, the framework continues. The product doesn’t go away. What closes is the window where your feedback can directly steer the roadmap. That window is sized to one solo developer’s bandwidth, which is why it has to close.
Founding cohort · Closes September 1, 2026
→ Join the Free CRAFT Beta