VUST Custom Solutions · Bounded AI Agents and Workflow Automation
Workflows that run on schedule — and wait for your approval.
For COOs · Operations directors · Department heads · Product owners
A recurring multi-step process becomes a bounded agent: explicit tools, finite authority, human checkpoints, repeatable recovery, and a log of every run with its cost. Not an autonomous employee — a state machine your team can see.
Explicit tools · Finite authority · Every run logged with its cost
Agent run log
Illustrative — not a client case
Trigger · Monday 07:00 · schedule
Weekly competitor summary: collect from 3 named sources, prepare the draft, request approval.
- Connected systemsSystemNamed APIs queried; inputs validated and versioned
- Bounded tool actionsSystemDraft prepared with explicit tools — no side effects yet
- Approval gateYour teamNamed owner reviews; nothing is sent before sign-off
- Output & recordSystemReport delivered; the run is recorded with provenance
- Retries & monitoringOwner + VUSTFailures retried by rule; every run logged with cost
The agent's authority ends where your approval begins.
01 · Fit
When is this the right fit?
The same inputs are collected every week
People pull from the same systems, apply the same rules, and paste the result together.
Report assembly eats days
The output is predictable; the collection and formatting are not automated.
The process lives in one person's head
Rules exist as habits, not written steps — vacations break the workflow.
Copy-paste between systems
Manual transfer is the integration layer, with all its typos.
Approvals happen by chat memory
Sign-off exists, but nothing enforces or records it.
Failures surface at the deadline
A silent breakage upstream is discovered when the output is due.
02 · Process transformation
What changes in the process?
Today
- Inputs collected by hand from several systems
- Rules applied from memory, differently each time
- Approval chased in chat, without a record
- No visibility into cost, failures, or history
With the system
- Scheduled collection from named systems with validation
- Drafts prepared by written rules an owner can change
- An approval queue with a named owner and a record
- Every run logged with cost, retries, and terminal states
Human boundary
Money, publication, hiring, and irreversible actions keep explicit human approval. The agent's tool set and authority are written and finite — that boundary is part of the acceptance matrix.
03 · Deliverables
What VUST delivers — and what is custom
Three honesty tiers: what VUST already operates in production, what is designed for your project, and what requires a discovery-stage audit before it can be promised.
Operating at VUST today
- Scheduled and background workflows with recovery
- Multi-step tool orchestration in VUST's own products
- Human-confirmed callback patterns
- Idempotent job and money-path patterns
- Usage accounting and per-run cost attribution
Designed for your project
- Workflow map and explicit state machine
- Tool set and authority limits
- Approval checkpoints and screens
- Exception and retry policy
- Evaluation set and acceptance matrix
Scoped after discovery
- Your system APIs and credentials
- Data warehouse and BI connections
- n8n hosting and ownership model
- Notification destinations
A named integration is treated as custom work until an access and contract audit verifies it.
04 · Example scopes
What a first release looks like
Three example scopes drawn from workflows VUST operates in its own products. They illustrate shape and boundary — not completed client cases.
Recurring research and reporting agent
Example scope — not a client case
A weekly analysis is assembled by hand from the same sources, formatted, and chased for sign-off.
- Buyer
- Operations or analytics lead
- Inputs
- Named sources, report template, approval rules
- Systems
- Source APIs and documents; delivery destination scoped per system
- VUST delivers
- Scheduled collection, draft preparation, approval queue, delivery, run log with cost
- Human gate
- The report ships only after named-person approval
- Measured target
- Hours from collection start to approved delivery, measured against your baseline
- Excluded
- New data sources mid-scope, ad-hoc analyses, autonomous distribution
- First engagement
- Discovery & architecture for one workflow
Support operations agent
Example scope — not a client case
Operators spend the first minutes of every case collecting context from three systems.
- Buyer
- Head of support or operations
- Inputs
- Case sources, enrichment rules, triage criteria
- Systems
- Ticket source; internal systems as custom integrations
- VUST delivers
- Triage and enrichment steps, prepared case bundle, exception queue, run monitoring
- Human gate
- Operators own every customer-facing action
- Measured target
- Case preparation time, measured over a fixed pilot window
- Excluded
- Auto-responses to customers, SLA promises, cross-team routing
- First engagement
- Discovery, then a focused MVP
Document preparation workflow
Example scope — not a client case
Incoming documents are read, extracted, and re-typed into the system of record by hand.
- Buyer
- Operations at a professional or finance-heavy firm
- Inputs
- Named document formats, extraction fields, validation rules
- Systems
- Document intake; system of record scoped as custom integration
- VUST delivers
- Extraction pipeline, validation, review queue, structured record, exception handling
- Human gate
- A person approves each record before it is written
- Measured target
- Share of documents passing validation on first review
- Excluded
- Unnamed formats, OCR of degraded scans without an audit, legal interpretation
- First engagement
- Discovery with a document-sample audit
05 · System anatomy
What is a bounded agent made of?
Six named parts. Each one has an owner and an acceptance check — this is what discovery maps for your workflow.
System
Trigger
Schedule or event — deduplicated, with a run identity
VUST
Workflow state machine
Explicit steps and terminal states; no silent loops
Custom scope
Tool layer
Named system actions with written limits
Your team
Approval checkpoint
Named owner; side effects wait here
System
Output & record
Delivery plus provenance of every input
Owner + VUST
Monitoring & recovery
Retries by rule, alerts, cost per run
Coral marks the human-authority component. Actor tags show who owns each part in operation.
06 · First engagement
How does a project begin?
A full production build is not VUST's default first step. The entry matches the state of your workflow.
Default
Paid discovery
Workflow map, system and data audit, authority design, architecture, MVP boundary, go/no-go.
Right when: The process is real and recurring, and the boundary needs defining before any build.
When prerequisites are finite
Focused MVP
One workflow, one KPI, one or two real integrations, one approval boundary, monitoring.
Right when: Discovery is done — or owner, KPI, tools, and approval design are already written.
Existing automation
Audit & stabilization
Read-first audit of the automation you already run: evidence, risks, stabilization plan.
Right when: An automation exists but duplicates actions, stalls, or costs more than anyone expected.
Discovery
Workflow, systems, authority, risks
Architecture
State machine, tools, approval design
Focused MVP
One workflow, one gate, acceptance
Production
Roles, data handling, rollback
Operations
Handoff or contracted monitoring
You can stop at any gate · Agents almost always enter through discovery — authority design is the hard part
07 · Cost and timing
What affects cost and timing?
Exact pricing is stated in your proposal after discovery — never as a public number. These are the factors that move it.
Price and schedule drivers
- Systems and API quality
- Data sources
- Run frequency and volume
- Approval design
- Exception complexity
- Roles
- Reporting
- Support model
Always separate lines
- Provider and model usage
- Hosting
- Third-party services and licences
- Taxes, where applicable
Timing starts when access, decisions, and prerequisites are complete. Consumer VUST pricing does not apply to B2B delivery.
08 · Why VUST
Why VUST for bounded agents?
Operating evidence — 07·2026
- Scheduled and background workflows running daily
- Multi-step tool orchestration in VUST's own products
- Human-confirmed callbacks and idempotent money paths
- Per-run cost attribution and structured logging
Facts refer to VUST's own products — not client cases.
After launch
Either an operational handoff — runbook, monitoring, a named owner on your side — or contracted managed operations with agreed coverage and reporting.
Workflows drift as systems change. The operating agreement includes a review cadence for rules, tools, and authority — so the agent's boundary stays true.
09 · FAQ
Clear before the contract.
Buying & scope
Is this an autonomous 'AI employee'?
No. VUST designs bounded multi-step workflows with explicit tools and human approval. Authority, recovery, and side effects are finite and written — 'autonomous employee' is exactly what this is not.
What is the smallest useful start?
Paid discovery for one workflow: map, systems and data audit, authority design, architecture, and a go/no-go. Agents almost never skip discovery — authority design is the hard part.
Implementation & data
Which systems can the agent touch?
Named systems only, after an access and contract audit of each. Actions with side effects are dry-run tested and gated behind approval until acceptance.
What happens when a run fails?
Retries follow written rules; every run reaches a terminal state; failures alert a named owner. Silent loops and stuck jobs are explicit anti-goals of the state machine.
Operation & ownership
Who approves the agent's output?
A named person on your side, through an approval queue that records the decision. The approval boundary is part of the acceptance matrix, not a UI afterthought.
What does a run cost?
Provider usage is metered per run and attributed. You see the cost history in the run log; estimates are labeled as estimates.
Next step
Describe the workflow your team repeats every week.
The brief takes about five minutes: the process, the systems it touches, volume, and the outcome you need. It ends with a concrete recommendation — discovery, a focused MVP, or an audit — prepared as guidance for the scoping conversation.
6 steps · Review before send · Nothing sent without your consent