Augment Group · AI operating systems for institutions

Intelligence,
augmented.

AI should amplify human judgment — not replace it. We build AI operating systems for the institutions that build the world: one governed memory of everything your organisation knows, with agents that act on it. Judgment that compounds.

01 · The shiftNot a cycle · a threshold

Every prior wave changed our tools. This one changes thinking itself.

Set it against the printing press and the industrial revolution — not against software cycles. Computers digitised work. The internet connected it. Mobile made it ubiquitous. Each changed industries; none changed the nature of work. Artificial intelligence does — one of the largest changes humanity has ever made.

1970s

Computers.

Filing cabinets became databases. Digitisation and efficiency — but work stayed manual.

1990s

The internet.

Connectivity and reach. Interactions became faster, global, always on.

2010s

Mobile.

The internet entered every pocket. Ubiquity became the default.

2020s →

Artificial intelligence.

Reasons, drafts, and decides — under human judgment. Different in kind, not in degree.

Bigger than the waves before it.
Bigger than all of them together.

02 · Watch it thinkOne question · every source · cited

See it think.

Questions that took a week of chasing resolve in minutes — drafted from your own record, cited to the line, verified before a human ever sees it.

Sources cited · verified · routed for review
03 · What we buildOne memory · one orchestrator · many agents

An operating system for your organisation's judgment.

Not another app, and not a chatbot over your files. An operating layer that turns everything your institution produces — decisions, documents, models, relationships — into one governed memory that agents read from and write back to.

One memory beneath them all. Every agent reads from and writes back to the same governed record — nothing learned at one desk is lost to another.

The orchestrator in the middle. Plans the question, retrieves across sources, cites the evidence — and routes anything below confidence to a person, context attached.

Agents per discipline. Diligence, contracts, close, pipeline, people, operations, reporting, dispatch — added sprint by sprint, never as a big bang.

Judgment stays human. Autonomy is earned per workflow and measured by evals — never assumed.

The memory

One memory.

Every decision, document, and relationship — including the paths not taken and the reasons why. The numbers connected to the reasoning. Nothing readable more broadly than its source.

The orchestrator

One round-trip.

Every question is planned, retrieved, fused across sources, and cited to the evidence — or the system says "I don't know." The same governed path, every time.

The agents

Many agents.

Investment and corporate, operational and administrative — drafting in minutes what took weeks. Humans review and decide. Autonomy is earned, never assumed.

Each decision makes the next one faster. That is the moat — and it cannot be bought.

05 · One systemFour worlds · one chassis

Four worlds. One operating system.

The doors are different; the house is the same. Every vertical runs the same chassis — one governed memory, one orchestrator, one agent runtime — instantiated against its own reality. What compounds in one compounds for all.

One memory One governance Many agents Institutional capitalSourcing → LP report EnterprisesHR · finance · ops · ERP Digital infrastructureNOC · field · assets SovereignsIn-country · owned
One memory · one governance · many agents
Institutional capital — sourcing → LP report
Enterprises — HR · finance · ops · ERP
Sovereigns — in-country · owned
Digital infrastructure — NOC · field · assets

An asset manager runs capital on it. Its operating companies run the field on it. Its corporate spine runs on it. A sovereign owns it outright. Same memory. Same governance. Compounding across all four.

06 · How we workNot a programme

Working software every 30 days. Not an 18-month promise.

The cadence

Ship every 30 days.

Working product in real hands every month, starting day 14. Course-correct on evidence, not opinion. If a sprint slips, scope is cut — never the timeline.

The team

A pod, not an army.

A lean senior team inside your walls, with your product owners holding the design pen. Knowledge transfer is designed in from day one.

The ownership

Yours, outright.

Every line of code, every prompt, every agent definition lives in your tenant. The moat compounds for you — not for a vendor.

07 · Proof, not promiseThe anchor engagement

A real institution runs on this. Today.

Our anchor partner — a permanent-capital infrastructure investment platform operating across three continents — went from kickoff to a board-approved AI operating system in eight weeks. Every employee was working in the new stack by day fourteen.

Day 14
Firmwide AI live — every employee in the new stack
8 weeks
From kickoff to board-approved architecture
−50%
Diligence cycle — 8–12 weeks to 4–6
+60%
Deal throughput — same team, deeper analysis
08 · The pathFirst conversation → first ship

From first conversation to first ship in four weeks.

Weeks 1–2

One-day diagnostic.

On-site assessment and executive workshop. You keep the gap analysis and target architecture either way.

Week 3

Phased proposal.

Sequence, scope, and team shape — a decision-ready case for your board or investment committee.

Week 4

Kickoff.

Sprint one begins. Engineers onboarded, foundations laid, first workflows instrumented.

Day 30

First demo.

Working software in front of your executive team. Course-correct on evidence, not opinion.