Give the agent the assignment
Message the agent to prepare a specific bid, or hand it the drawing package directly. The request describes the outcome instead of the screens to operate.
← All case studies
Agent-directed estimating · Live foundation
Ask for the bid. Review the result.
The destination is not another quoting tool someone has to operate screen by screen. The estimator messages an agent—or gives it a drawing package. With explicit authorization, the agent gathers the source material, operates the quoting engine, prepares the working bid and quotation drafts, groups the exceptions, and returns everything for human review and approval.
See how the system works ↓Independent illustration · target experience
R0
Estimator agentPrivate workspace
ActiveYou
Check the new bid request in my authorized inbox and prepare the complete draft.
Agent runPreparing review package
✓
Source package collectedAuthorized inbox + drawing set
✓
Documents indexedSources preserved
✓
Takeoff candidates preparedEvidence attached
✓
Pricing coverage checkedControlled cost sources
!
Exceptions grouped7 decisions need review
Agent
Draft working bid ready
I prepared the bid and quotation drafts and grouped the exceptions. Nothing has been approved or sent.
Review draft
Public-safe representationNo client data, screens, records, or proprietary source.
StatusLive engine · Agent workflow in development
RoleProduct lead + builder
ScopeAuthorized intake through review
OutputDraft bid + review package
01 The operational problem
The opportunity is to remove the software-operating burden—not the estimator’s judgment.
Most quoting software still asks an estimator to navigate screens, upload and organize documents, clean up extracted data, check coverage, resolve scope, assemble pricing, and generate the final package. The software may be better organized, but the person still operates every step.
The target experience is outcome-directed: tell an agent what bid to prepare, or provide the PDF, and let it operate the underlying system. That only works if every action stays tied to authorized sources, evidence, visible exceptions, and explicit human approval before anything becomes final or goes to a customer.
02 The system
Instruct the agent. Let it prepare the work. Review the complete draft.
This is the product direction. The current private pilot is the controlled quoting engine underneath it; the full agent-operated experience is still being built.
Gather authorized inputs
With explicit permission, retrieve the relevant email, drawings, specifications, supplier packages, and reference material from approved sources only.
Operate the quoting engine
Create the project, organize the documents, prepare source-backed takeoff candidates, and run the system’s controlled preparation tools.
Prepare the bid
Build the draft quantities, pricing coverage, scope, allowances, and exclusions while preserving evidence and surfacing uncertainty.
Assemble the review package
Return a working bid, customer quotation draft, source trail, and a grouped list of the decisions that still need estimator attention.
Human review and approval
The estimator reviews, edits, and approves the result. The agent does not silently approve a quantity, choose an assumption, or send a quotation.
03 What I built
Product thinking, engineering, and operational controls in one delivery.
Agent-first product direction
Designed the experience around an outcome-level instruction—prepare this bid—rather than making the estimator manually operate every stage of another software product.
Agent workflowProduct designEstimator experience
Controlled quoting engine
Built the underlying application, data model, APIs, project history, review states, evidence trail, readiness controls, and PDF/XLSX export pipeline the agent can operate.
Full stackStructured dataControlled outputs
Agent authority boundary
Defined authorized inputs, auditable actions, visible exceptions, and human-only decisions so automation can prepare the work without becoming an unaccountable approval authority.
Agent safetyHuman approvalAuditability
Private delivery and QA
Built local and private-cloud delivery paths, guarded releases, persistent storage, backup and restore, and automated unit, integration, browser, and deployment verification.
Private cloudRelease engineeringAutomated QA
04 Current stage
The quoting engine is live. The agent-operated experience is the direction now being built.
The private pilot provides the durable project record, deterministic authorities, evidence trail, review controls, and export gates the agent will use. The next layer changes how the estimator delegates the work: from operating the application directly to assigning the outcome and reviewing what comes back.
- 01The underlying private quoting engine is deployed with authenticated access and persistent project data.
- 02Drawing intake, reviewable takeoff, pricing coverage, scope blockers, and controlled export paths are implemented.
- 03The agent workflow is designed to invoke those existing authorities rather than calculate or approve work independently.
- 04Authorized email retrieval, end-to-end agent orchestration, and estimator acceptance remain in development; they are not presented here as complete.
05 Public boundary
Specific about the work. Careful with the client.
Included
The operational problem, agent-first product direction, human approval boundary, my delivery role, implemented quoting foundation, current stage, and a newly created synthetic illustration of the target experience.
Excluded
Client identity, project drawings, customer or supplier information, pricing, proprietary cost data, production screenshots, source code, credentials, internal metrics, and acceptance statements.
R0
06 Start with the workflow
Have an operational process that still depends on chasing, re-entry, and individual memory?
Routine Zero can map the work, identify the real control points, and build the smallest dependable improvement around it.
Tell me about one workflow →