HOLD.co

Technology · HOLD.co portfolio

vb.co — Operating AI, the layer between a company’s systems and how it actually works.

vb.co builds Operating AI: one intelligent layer connecting a company’s people, workflows, data, tools and decisions. Sixteen department and workflow agents, an integration layer over the systems already in place, and governance on every action — delivered through a phased method that begins with an assessment rather than with software.

A HOLD.co company

An operating layer built by an operator.

vb.co is a HOLD.co technology company, sitting alongside DEV.co, ERP.io, LLM.co, Automatic.co, RMA.ai, Search.co, RFP.co, VDR.ai, Manufacturing.co, Phony.ai and ProjectManager.co in our technology segment.

The thesis comes from running the portfolio. HOLD.co owns operating businesses in asset-heavy industries, and in every one of them the same pattern shows up: the systems are adequate, the people are good, and an enormous amount of the week goes into moving information between the two by hand.

Operating AI is the answer to that, and it is deliberately not another application to buy. It sits over the systems a company already runs, which is the only version of this that a middle-market business can actually adopt — replacing the stack was never on the table.

The problem

Four symptoms, and none of them look like an AI problem.

Companies rarely arrive asking for agents. They arrive with a set of ordinary operational complaints that have one thing in common: the work is repeatable and a person is still doing it.

Someone rebuilds the same report every Monday

The work is repeatable, well understood, and done by hand anyway, because the data sits in four systems that were never asked to talk to each other. It is the cheapest work in the company to automate and the last thing anyone gets to.

A process lives in one person's head

It runs beautifully until that person is on holiday, and then it stalls in a way nobody can diagnose because it was never written down. Institutional knowledge that exists in exactly one place is a liability wearing the costume of an asset.

Exceptions surface when a customer complains

The threshold was crossed days earlier and nothing watched it. By the time the problem is visible it is a relationship issue rather than an operational one, which is a far more expensive category of problem.

Nobody can reconstruct why an approval went through

The decision was made, the money moved, and the reasoning is gone. That is uncomfortable in an audit and genuinely dangerous the first time a regulator, an acquirer or an insurer asks the question.

The platform

Agents, workflows, knowledge and governance.

The four pillars are deliberately inseparable. Agents without governance are a liability, workflows without knowledge produce confident nonsense, and knowledge without agents is a search box nobody opens.

Agents

Specialized AI workers that execute repeatable tasks against your own systems rather than in a chat window beside them.

Workflows

Processes that run without a person coordinating each handoff, and escalate when something falls outside the rule.

Knowledge

The company's operating knowledge indexed and searchable, instead of resident in a handful of people and a shared drive.

Governance

Oversight, controls and an audit trail on every action — every action carrying its actor, its input, and the rule that authorized it.

The roster

Sixteen agents, organised the way a company is.

Some map to a department and some to a workflow that crosses several. Each is scoped to work a real desk rather than to be a general assistant, because the narrow ones are the only ones anyone can actually hold accountable.

Executive
Sales
Marketing
Customer support
Operations
Finance
HR
IT
Proposal
Invoice
Recruiting
Knowledge
Reporting
Procurement
Compliance
Meeting

How it lands

Assess, blueprint, prototype, roll out.

The reason most AI programmes fail is not the models. It is that they start at the software and work backwards to a business case, which produces a pilot nobody can extend and a result nobody agreed how to measure.

01

Assess

An AI readiness assessment across your data, systems and processes, plus a strategy workshop that produces an opportunity map and an executive alignment document. The output is a ranked list of where this is worth doing, including where it is not.

02

Blueprint

A system integration blueprint and a phased roadmap, with a success metrics framework agreed before anything is built. The measurement is defined while it can still be argued with, rather than assembled afterwards to fit the result.

03

Prototype

A prototype sprint that deploys working agents against real systems and real data, with the integration architecture and a first governance framework in place. Judgment happens against something running, not a deck.

04

Roll out

Full deployment department by department, each one carrying its own controls and its own measures, so the programme compounds rather than arriving as a single cutover nobody can reverse.

Who it serves

SMB and middle-market companies with real operations.

The pattern holds across all of them: enough process to be worth automating, enough complexity that the process is scattered, and not enough headcount to keep doing it by hand.

Manufacturing
Construction
Logistics
Healthcare
Professional services
Financial services
Real estate
Retail and ecommerce
Energy
Field services
Distribution
SaaS and technology

Where it doesn’t fit

Three honest boundaries.

It is not a product you buy on a Tuesday and switch on. Operating AI arrives through an assessment and a phased rollout, which is a commitment of attention from the leadership team as much as a budget line. A company that wants software delivered and nothing else should say so early.

It is not a replacement for the systems underneath it. The ERP, the CRM and the accounting package stay exactly where they are — the operating layer reads and writes across them, and a company hoping to retire one of them with this is solving a different problem.

It does not remove the need to decide who is accountable. Governance is enforced in the software, but somebody still has to say which approvals an agent may make and which it must route to a person. That work is not technical, and it is the part that determines whether any of it holds up.

Our thesis

The middle market will not replace its systems to adopt AI, and every product that requires it will be evaluated and declined. The layer that wins is the one that treats the existing stack as permanent and makes it answer questions it was never able to answer before.

Start with the assessment, not the software.

The readiness assessment produces a ranked map of where this is worth doing in your business — including the places it is not, which is usually the more useful half. Book a walkthrough, or put your own numbers through the savings calculator first.

Interested in the platform or the portfolio?

Whether you want an operating layer over your own systems or you're an owner, operator, or investor exploring a deal with HOLD.co, start a confidential conversation.