Skip to content
VegaStack home
Book a demo

AI agents for the work between tenders and quotes, purchase orders and the ERP, specs and submittals, invoices and payments, tickets and releases.

We build agents into the systems you already run (your ERP, your document control, your repositories) for estimation, order entry, finance operations and software delivery. They read the documents, do the matching and drafting, and leave your people the decisions. We agree the metric before we build, and you own the code.

Book a demoHow an engagement runs

170+ automation projects · 65 companies · 6 years

Illustrative run: BOQ lines from a hospital lighting tender matched to catalogue items. Most match with high confidence, one is flagged for a 2.5% lumen shortfall, one names another make and gets a declared equivalent, and one bollard is left unmatched because no item meets the height.
Illustrative run: a customer purchase order of seven lines read from a PDF. Six lines map cleanly to the item master; one has a revision mismatch and a price 2% below contract, and the requested date is two days before availability. The agent drafts the order for six lines and an email to the buyer about the seventh.
Illustrative run: a lighting submittal compliance statement. Each specification clause is compared with the datasheet value and cited to its page. Three clauses comply, one efficacy value is declared as a deviation, and one certificate expires before delivery and needs renewing.
Illustrative run: a supplier invoice for 200 units against a purchase order for 200 and a goods receipt for 180. The agent holds the 20 unreceived units, confirms the GST fields, suggests the GL account and cost centre from past invoices, and drafts a request for a credit note.
Illustrative run: a pull request opened by an agent from an approved plan. It rounds GST per line instead of on the total. Types, tests, security scans and a second-agent review have passed; it is waiting for the required human code-owner review before anyone can merge it.

Six years of automation work for teams at

Point of view

Most AI pilots stall for reasons that have nothing to do with the model. We design every engagement around these four.

Why pilots stall

  • 55%

    of enterprises rank integration among the top reasons pilots fail to scale.

    Madrona, 2026
  • 52%

    rank security, privacy and compliance; it's also the first thing buyers check.

    Madrona, 2026
  • 40%+

    of agentic AI projects will be cancelled by 2027 over cost and unclear value.

    Gartner, 2025
  • 70%

    of agentic AI built by vendor engineers will be abandoned by 2028 when nobody inside can change it.

    Gartner, 2026

Where we work

Built desk by desk, in the words your team already uses. Pick your industry.

Every workflow below is something we build: what comes in, what the agent does, who approves, and the number we agree to move.

For contractors, EPCs, MEP subcontractors, consultancies and building-products manufacturers, we build agents for the document work between a tender and a signed contract: pricing BOQs against your rates, briefing tender packs, levelling subcontractor quotes, building compliance statements and keeping notices on time. They work in Excel, Aconex, Procore and your ERP, and every price, deviation and notice is approved by a person.

Preconstruction and tendering

Technical office

Commercial

  • Contracts, notices and claimsObligations and time-bars are pulled into a register, letters and site records are linked to the clauses they touch, and notices are drafted before the deadline for the commercial manager to sign.MetricNotices issued on time
  • Variations and change ordersRevised drawings and instructions are compared with the contract BOQ, and the variation is drafted with quantities, rates and the instruction it came from.MetricDays from instruction to priced variation
Construction and engineering: the full picture

Anatomy of a run

What a production agent actually does. One priced BOQ, start to finish.

  1. 01It reads the document as it arrivesThe consultant's sheet, with its merged cells and sub-items, not a copy pasted into our template. Every line keeps its source row.
  2. 02It checks the document against itselfBOQs disagree with schedules and specs more often than anyone likes. Those lines are flagged before anything is priced.
  3. 03Rules first, then judgementYour catalogue is filtered with plain rules before a model ranks the shortlist against a written rubric. That keeps cost down and every rejection explainable.
  4. 04It's allowed to say noA wrong match is a failure; “no match” is an acceptable answer. Uncertain lines go to the estimator with the reason.
  5. 05It stops for a personAnything outside authority, like a price under the margin floor, waits for the person who owns that decision. Waiting costs nothing.
  6. 06It writes back, and keeps the recordThe priced BOQ goes out in the client's format, with every step, override and approval in the log.
Illustrative run log for a priced BOQ: the agent reads the sheet, reconciles it with the schedule, extracts attributes, filters the catalogue, matches and prices the lines, then asks the sales head to approve three lines under the margin floor. It waits without running until a person approves, then exports the priced BOQ, compliance sheet and deviation register.

Controls in code

The controls live in the code, not in a prompt. These are what your security team will ask about first.

Security and integration are the two things buyers check first and the two things that kill pilots. Each control below is built in and shows up in the logs.

  • It can say “no match”

    When the evidence isn't there, the agent abstains and tells you why, instead of filling the gap with a plausible guess.

  • Every statement cites its source

    Answers, matches and flags link to the page, clause or row they came from. What can't be found is dropped before anyone sees it.

  • People approve what goes out

    Prices, orders, payments, notices and merges are approved by the people who hold that authority today. The run waits; nothing is spent while it does.

  • Everything is in the log

    Every read, decision, override and approval is recorded with who or what did it, so an audit is a query, not an archaeology project.

  • Risky actions are fenced

    Agents get only the permissions the step needs. Payments, deletions and merges are outside what an agent can do, and paid calls are never repeated by accident.

  • It runs where you choose

    Your cloud account or ours, in the region your data has to stay in, with model providers that don't train on your data.

How an engagement runs

Start with a demo on your own documents. Pay for production only when the numbers say so.

  1. 01 · 1 hour, then 1–3 days

    Demo on your documents

    Send 10 to 20 real documents: a BOQ, a stack of POs, last month's invoices, a backlog of tickets. We run an agent on them and walk you through what it matched, what it flagged and what it refused to guess.

  2. 02 · 2 weeks

    Workflow assessment

    We sit with the people who do the work, measure how it runs today and check what the agent would need to read and write. You get a shortlist of workflows with a scored case for each, and a fixed quote for the first one.

  3. 03 · 6–8 weeks

    First workflow in production

    One workflow, live in your systems, on real work. We build the test set from your history first, so accuracy is measured before anyone relies on it, and part of the fee is held against the metric we agreed.

  4. 04 · 4–6 weeks each

    Next workflow

    Each further workflow reuses the integrations, test tooling and approval patterns from the first, so it's faster and costs less.

  5. 05 · Monthly

    Run and improve

    We keep the agents healthy: monitoring, model and prompt updates, the test set re-run on every change, a review of the exception queue and a quarterly review of value against the baseline.

What you own at the end

Agents get abandoned when nobody inside the company can change them. So handover isn't an extra; it's in every engagement.

Durations, deliverables and what we need from you
  • The code, prompts and configuration, in your repository
  • The test set built from your work, and the results
  • Runbooks for operating and changing the agents
  • Dashboards for the metric we agreed
  • Integrations, under accounts you control

Where the agent sits

No new app for your team to learn. The agent works where they already do.

It connects through the same APIs and validations a person's entry would hit, under accounts you control, and hands every decision to the person who owns it. Start read-only, then draft-only, then let clean cases through.

Your systems

  • SAP
  • Dynamics 365
  • Oracle
  • Zoho
  • Excel
  • Outlook
  • SharePoint
  • WhatsApp
  • Salesforce
  • GitHub
  • Jira
  • Slack
  • Tally
  • Aconex
  • Procore
  • your TMS

The agent

  1. read
  2. check against your rules
  3. draft
  4. ask a person
  5. write back
  6. log

Your people decide

  • Estimator
  • Sales head
  • Customer service
  • AP specialist
  • Engineer
  • Code owner
The agent sits between your systems and your people: it reads from and writes back to the systems you already use, and every decision that matters goes to a person.

Selected work

What we've built, described plainly. Clients are described rather than named until they sign off.

Status words mean what they say. Shipped is in use, in client testing is not yet in production, and in daily use is how we run ourselves.

Questions

What buyers ask us first.

Send us ten real documents. We'll show you what an agent does with them.

A one-hour call with a founder and an engineer. Bring a BOQ, a stack of POs, last month's invoices or a backlog, and leave knowing what would work, what wouldn't, and what it would take.

Book a demoor email founders@vegastack.com

170+ projects · 65 companies · founders on every engagement