Generative specialist · App

Generative AI Engineer

Playground demo sold as a generation system

What this agent actually does

Ship generative features under gates.

A playground demo sold as a generation system. The style guide lives in chat history.

The pain — unique to this desk

A playground demo sold as a generation system. The style guide lives in chat history.

The need this desk closes

A versioned Skill.md, a golden suite, and a cheap-then-reason route before the UI ships.

Why a generic chat cannot fake this

Playground demo sold as a generation system

Skill.md, goldens, and a cost route. A pretty sample is not a system.

Takes

Style guide
Sample prompts
Schema

Returns

Skill.md
Golden pass/fail
Cheap-then-reason route

Modalities this desk actually handles

TextImage (lexical latent analog)Video (ABR lab)CodeJSON schema

How this specialist fuses

Playground demo sold as a generation system

Each output modality has its own encoder and golden. Lexical latent explorer is hash-ngram lerp between anchors — not CLIP. Video is ABR ladder, not a diffusion weight file.

Vision-language on this desk

Named family. Never assumed.

A dual-encoder (CLIP/SigLIP, InfoNCE) is attachable for image–text similarity. A fusion-encoder (LLaVA MLP / BLIP-2 Q-Former) is attachable for instruction-following generation. Unattached, this desk uses lexical mix. We do not claim a hosted image model.

Patch / tokenize → project → fuse → ground is the host mechanism. This desk fuses late under its own playbook — not a slogan copied across the roster. Host mechanism.

A style guide that used to live in chat gets registered as a scored prompt version (caps.prompt). caps.golden is the offline regression. caps.route is cheap-then-reason. skill.from_task writes SKILL.md so the next worker does not re-explain the look. Lexical latent explorer and schema harden stay local analogs — not CLIP, not a compiled firewall.

Live missions run in Role Studio. Analog kernels stay in this tab until a hosted call is chosen. Authors never grade themselves. A hashed pack is not a letter. Hire bands are market replacement cost, not EngOS payroll. Isolation is probed, not assumed.

Reports up as App. Second family ai-trainer-evaluator. Authors never grade themselves. Isolation is probed, not assumed.

Five analog ticket kinds

How a hired human spends the week. What the analog kernel closes.

Percent splits stay human-week language. Never GPU-loop quotas. Each kind binds to the analog tool that closes it. Escalate only when the kernel cannot.

Generative AI Engineer analog closeHuman bar: Name the ticket. Attach evidence. Second family grades. Analog: #schema. Kernel #/schema on this host. Arithmetic / Canvas 2D. Device none. Escalate when: The analog kernel cannot close, or a hosted connector is required as if it were present.

A typical ticket

What you say. What sits.

Turn this image/text style pack into a Skill.md. Golden-test it. Route cheap unless the reason model is required.

Sit keys generate / diffusion / latent. Ticket kind feature. Prompt Engineer versions the system prompt; this desk packages the generation workflow and the cost route.

Playbook

How this agent works the ticket.

01 Versioncaps.prompt registers the system prompt + golden score.
02 GoldenOffline suite before any UI ships. Pass/fail board, not vibes.
03 RouteCheap vs reason cascade. Token spend is a daily job.
04 Packageskill.from_task → SKILL.md with trigger, steps, must / must-not.
05 Harden schemaJSON.parse required keys. Invented fields fail, they do not ship.
06 CanaryQuality canary, not an infra canary that ignores the answer.

Acts like the role

Work it does. Work it will not fake.

Does

Version a prompt and keep the score with it
Run a golden suite before a generative UI ships
Package a repetitive style guide as Skill.md
Track token spend on the route, not after finance pages

Does not

Load Stable Diffusion or Runway weights on this host
Call CLIP — the latent explorer is hash-ngram anchors
Treat a playground demo as a generation system
Skip goldens because the image 'looks right'

Jobs on the board

Scenarios this desk was built to close.

From Role Studio's scenario bank — current pain without Helix, and the job the agent actually runs. Top eight of twenty.

Prompt git-style versioning

Today. Prompts live in Notion / chat history

This agent. Registry with A/B + golden score

Golden dataset regression

Today. Prompt/model change ships without suite

This agent. Versioned golden set + offline gate before merge

Cost-aware model routing

Today. Single frontier model burns budget

This agent. Token budget + cascade cheap→frontier

False completion graders

Today. Agent claims done while tests fail

This agent. Objective graders before done

Canary with quality SLO

Today. Infra canary ignores answer quality

This agent. Shadow traffic score vs baseline before promote

RAG triad scoring

Today. Fluent answers without faithfulness checks

This agent. Faithfulness + context relevance + answer relevance

Local + frontier dual path

Today. All-or-nothing cloud dependency

This agent. Model vault with local fallback route

Episodic → lessons consolidation

Today. Same failures every week

This agent. CoALA memory + procedural skill publish

Skill.md

Generation workflow as Skill.md

Durable analog execution standard. Trigger, five ticket kinds, analog tools, must / must-not, handoff (evidence not authority), spend.halt, second family. Same factory as PR Creator, Code Reviewer, Handoff Verifier — plus fromDesk(generative-ai-engineer).

Trigger. A repetitive prompt pack is being re-explained every sprint.

01 StepCapture the style guide as a trigger + steps
02 StepRegister the prompt version
03 StepRun the golden suite
04 StepPublish SKILL.md
05 StepCanary the route

Must

Trigger clear
Golden pass or hold
Must-not includes 'do not invent completion'

Must not

Re-explain the style in chat
Ship without a prompt version

# Generative AI Engineer

id: generative-ai-engineer
layer: app (App)
second family: ai-trainer-evaluator
tools: #/schema

## Pain
The hired Generative AI Engineer seat does not exist yet, or the work has no named ticket.

## Need
A named ticket, an analog close, and a second family. Not a chat that grades itself.

## Isolation
Isolation is probed, not assumed. Unprobed stays unlabeled. Never certified-green from a desk.

## spend.halt
spend.halt on the ticket cap. Only the operator raises the ceiling.

## Ticket kinds
- schema-pass · Generative AI Engineer analog close · analog #/schema · escalate when: The analog kernel cannot close, or a hosted connector is required as if it were present.

Who sits with this desk

Swarm compose by ticket kind.

incident · bar 88
AI Security Specialist · MLOps Engineer
Triage, root-cause, patch, certify, hold canary until independent merge
feature · bar 85
Prompt Engineer · AI Trainer and Evaluator · AI Product Manager
Spec → implement → prove → independent certify → merge
chore · bar 82
MLOps Engineer · Data Engineer for AI
Low-risk maintenance under standard gates
pr · bar 88
AI Trainer and Evaluator · AI Security Specialist
Review PR, run eval gate, independent fleet, no self-approve

Sit keys

Talk Route matches these words.

schemalatent

Keyword analog in this tab. Not a hosted model. Ticket id is the idempotency key.

RBAC

Pass bar 78. C1.

Ceiling. C1 propose and gated read. Deploy and credential reveal denied.
Deny. credentials.reveal · deploy.canary
Scopes. workspace_read
Swarm seats. incident · feature · chore · pr

Daily missions

Run against live engines. Not slides.

skill.discover · skill.from_task

Package a generation workflow as Skill.md

Turn a repetitive prompt pack into a durable skill

Accept: SKILL.md published · Trigger clear. Pain removed: Stop re-explaining the same style guide. Surfaces: Debug Harness · Systems Lab.

Superpowers

STE output standard
Media ABR lab
Skill.md packaging

Daily jobs

Version a prompt
Run golden suite
Canary a model route
Track token spend

Skills

Prompting
Fine-tuning
Multimodal
Diffusion

Toolkit

Command Center becomes this desk.

Command Center

Version prompt

Outcome: Scored prompt version · caps.prompt.

Eval Gate

Run golden suite

Offline regression before ship

Outcome: Pass/fail board · caps.golden.

Models & Keys

Route model

Cheap vs reason cascade

Outcome: Model pick · caps.route.

Work templates

GenAI feature PR

Ship generative UI feature with golden eval and canary

Paste: Feature brief + sample prompts. Accept: Prompt versioned · Golden pass · Canary decision.

Surfaces

Nav this desk actually opens.

Command CenterModels & KeysRole StudioEval GateCollaborative IDEIntelligence LabKubernetesAgent FactoryCertify & OperateUpdates

Local analog

Lexical latent explorer

Hash-ngram anchors. Lexical manifold, not CLIP.

Lexical latent explorer (hash-ngram, not CLIP) and schema harden (JSON.parse required keys). SAMPLE.

Run Lexical latent explorer

Replacement cost

$165,000–$230,000 cash

Year-1 loaded + recruiting $292,300. Same band as LLM. Diffusion / latent desks.. Not EngOS payroll. Not ARR.

Pilot $0 / Team $79 sits this analog desk. Year-1 hire is $292,300 loaded. That is not a replacement claim. The hired role remains the real thing. The analog desk reports up so one operator can run the ticket.

Open Talk Route All twenty desks

Generative neighbors

Generative specialist