The pain — unique to this desk
Agentic specialist · App
Forward Deployed Engineer
Client secrets in a chat that is not a tenant
What this agent actually does
Customer production AI.
Client secrets in a chat that is not a tenant. A go-live with no runbook and no citations from their docs.
The need this desk closes
Connect their data, put guardrails on it, wire the workflow. Scratch stays local — not a tenant.
Why a generic chat cannot fake this
Client secrets in a chat that is not a tenant
Connect their data, put guardrails on it, wire the workflow. Scratch stays local. A demo in the room is not a go-live.
Takes
- Customer context + symptom
- Corpus notes
- Screenshots
Returns
- Runbook
- Cited corpus answer
- Go-live checklist
Modalities this desk actually handles
TextCustomer docsImages as attachmentsAttachable connectors
How this specialist fuses
Client secrets in a chat that is not a tenant
Customer attachments (docs, screenshots, traces) land on the ticket. CV encodes images; RAG encodes docs; this desk wires the workflow. Scratch analog is not a tenant.
Vision-language on this desk
Named family. Never assumed.
If the client has a vision-language model, it attaches. Missing never fails the product. Isolation is probed, not assumed.
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 customer feature is down or a go-live is this week. The triad is named: connect data, guardrails, workflows. work jobs carry severity. rag.grounded answers from the customer corpus with citations. findings.fix_all is the go-live checklist (billing, webhook, deploy). Discovery scratchpad is the local analog — notes export as markdown. Client secrets do not live in a chat that is not a tenant. Palantir-style: most of the day is IC work in the customer's system, not a slide.
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.
- Forward Deployed Engineer analog closeHuman bar: Name the ticket. Attach evidence. Second family grades. Analog: #scratch. Kernel #/scratch 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.
Plant FAQ bot is citing a retired SKU. Triage, ground the corpus, patch, canary. Do not put their secrets in this tab.
Sit keys customer / onsite / forward / scratch / FDE. Ticket kinds support, incident, billing. PM prioritizes. Prompt Engineer tightens replies. Security reviews the embed. This desk is on-site analog, not a SOC letter.
Playbook
How this agent works the ticket.
- 01 TriageIncident job, severity-tagged, rollback path named.
- 02 Connect dataCustomer corpus in. Grounded answers with citations. Not a demo dataset.
- 03 GuardrailsSix-layer guard on their untrusted input. Secrets stay out of artifacts.
- 04 WorkflowsThe feature is a ticket loop with eval, not a chatbot in a slide.
- 05 Go-live chainProd-readiness fix-all: billing · webhook · deploy. Canary only after eval.
- 06 Leave the runbookLessons.md + scratch export. Next week does not start from zero.
Acts like the role
Work it does. Work it will not fake.
Does
- Triage a customer incident with a severity and a rollback
- Answer from their docs with citations
- Run the go-live checklist before a customer canary
- Keep discovery notes in a local scratchpad, not a fake tenant
Does not
- Turn this sales host into a client tenant
- Store customer secrets in chat
- Claim a Customer VPC or their K8s as live here
- Skip eval because the demo in the room looked good
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.
P1
Success metrics on AI features
Today. Ship AI without measurable outcomes
This agent. Feature brief + eval plan + KPI log
P1
Incident runbook agent
Today. Pager chaos without structured triage
This agent. Incident desk with severity + rollback path
P0
Agent loop with durable checkpoints
Today. Agents lose state mid-run; restart from scratch
This agent. Checkpointed harness with resume + budget per cycle
P0
Golden dataset regression
Today. Prompt/model change ships without suite
This agent. Versioned golden set + offline gate before merge
P0
RAG triad scoring
Today. Fluent answers without faithfulness checks
This agent. Faithfulness + context relevance + answer relevance
P0
Prompt git-style versioning
Today. Prompts live in Notion / chat history
This agent. Registry with A/B + golden score
P0
Canary with quality SLO
Today. Infra canary ignores answer quality
This agent. Shadow traffic score vs baseline before promote
P1
Cost-aware model routing
Today. Single frontier model burns budget
This agent. Token budget + cascade cheap→frontier
Skill.md
Customer incident
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(forward-deployed-engineer).
Trigger. A customer AI feature is down or a go-live is blocked.
- 01 StepTake customer context + symptom
- 02 StepTriage with severity
- 03 StepGround the corpus answer
- 04 StepPatch under eval
- 05 StepCanary or hold with a deploy decision
Must
- Triage
- Fix artifact
- Deploy decision
- No secrets in the scratch analog
Must not
- Paste client secrets into a non-tenant chat
- Courtesy-pass a go-live
# Forward Deployed Engineer id: forward-deployed-engineer layer: app (App) second family: ai-trainer-evaluator tools: #/scratch ## Pain The hired Forward Deployed 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 - scratch-pass · Forward Deployed Engineer analog close · analog #/scratch · 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.
support · bar 80
AI Product Manager · Prompt Engineer
Deflect with grounded corpus, escalate billing-sensitive cases
billing · bar 82
MLOps Engineer · AI Product Manager
Cost caps, plan, usage ledger — admin only paths
FDE triad from the product brief: connect data, guardrails, workflows. 80/20 IC / client. Not a Palantir clone claim — the work pattern is what we built.
Sit keys
Talk Route matches these words.
scratch
Keyword analog in this tab. Not a hosted model. Ticket id is the idempotency key.
RBAC
Pass bar 80. C2.
- Ceiling. C2 gated act. Eval bar required. Secrets stay denied unless the profile lists them.
- Deny. Empty denylist on this profile. Isolation is probed before it is certified.
- Scopes. workspace_read · workspace_write · corpus_read · corpus_write · queue
- Swarm seats. support · billing
Daily missions
Run against live engines. Not slides.
deploy.topology · sandbox.layers
Choose edge vs serverless for a client
Simulate deploy topology with cold-start / data residency tradeoffs
Accept: Topology recommendation with constraints. Pain removed: Architecture review in minutes not decks. Surfaces: Kubernetes · Systems Lab.
Superpowers
- Sandbox 3-layer stack
- Serverless / on-prem / edge topology picker
- Hardening checklist
Daily jobs
- Triage customer incident
- Ship fix under eval
- Update runbook lessons
- Run prod-ready fix-all before customer canary
Skills
- Deployment
- Integration
- Client systems
- Security
Toolkit
Command Center becomes this desk.
Command Center
Incident job
Severity-tagged recovery path
Outcome: Runbook artifact.
Intelligence Lab
Customer corpus answer
Grounded reply from their docs
Outcome: Citations.
Kernel Guard
Prod readiness
Billing · webhook · deploy fix-all chains
Outcome: Customer go-live checklist · findings.fix_all.
Kernel Guard
Consolidate
Architecture consolidation prompt + live run
Outcome: Arch score for customer stack · prompts.consolidate.
Work templates
incident
Customer incident
Customer AI feature down — triage, patch, canary
Paste: Customer context + symptom. Accept: Triage · Fix artifact · Deploy decision.
Surfaces
Nav this desk actually opens.
Command CenterModels & KeysRole StudioCollaborative IDEKubernetesEval GateIntelligence LabAgent FleetKernel GuardCertify & OperateUpdates
Local analog
Discovery scratchpad
Local notes. Export markdown. Not a client tenant.
Discovery scratchpad: local notes, export markdown. Not a client tenant. SAMPLE.
Replacement cost
$170,000–$250,000 cash
Year-1 loaded + recruiting $310,800. On-site triad. Palantir-style premium.. Not EngOS payroll. Not ARR.
Pilot $0 / Team $79 sits this analog desk. Year-1 hire is $310,800 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.
Agentic neighbors
Agentic specialist
AI Agent Engineer
Agent said done with no contract
An agent that said done without a contract. Weak state between workers. Courtesy pass on half a swarm.
Stamp → prove → certify. Handoff verified against the workspace. Self-report is not done.
TextTool tracesCode
- Certify a new support triager before it sits a ticket
- Debug a flaky handoff with the harness matrix
- Run a 3-agent graph on a customer incident
Agentic specialist
AI Automation Specialist
Trigger that pages the wrong person
A trigger that pages the wrong person, or never fires. The next agent sits and the job has to be re-explained.
Default-deny, human gate, Seat Token. A Zap is not a fleet.
EventsText
- Wire a safe trigger with a human gate on send
- Deny high-risk tools by default
- Process the exception queue
Agentic specialist
AI Engineer
Ten engineers for ten loops
Ten engineers for ten loops. A feature ticket with no eval and no canary decision.
Ticket → artifact → eval → canary. Fleet playbook under a constitution. One engineer covers the loops.
CodeTextAPIs
- Implement a ticket with agent assist and an artifact
- Gate a PR through eval
- Canary to staging with a probe trail