What it does

One machine. The team changes, not the software.

You are not buying a research tool, or a writing tool, or a security tool. You describe a goal; the system forges the specialists that goal needs, writes the plan, and runs it under your limits. The work below is nine kinds of work we have used it for ourselves, and none of it was built in: anyone, from any industry, can create the specialists and plans their own domain needs.

Why it composes

Three things are decided per goal, not per product.

The team

Specialists are forged, not chosen from a list.

A domain brief goes into the foundry and a full specialist comes out, with an oath written for that field. There is no catalogue to be missing from. If a domain can be briefed, it can be staffed.

Proof · 16 plans whose only job was staffing a domain

The plan

The plan is written for the goal.

Phases, steps, dependencies and the gates where a person signs are composed for the job in front of it, not selected from a template library and then argued with.

Proof · 177 plans · 158 distinct goals

The output

The artifact follows the work.

Some work ends in a document, some in a working application, some in a page. The same run decides which, and renders it in your brand rather than handing back a wall of text.

Proof · 488 documents · 308 apps and pages on the boxes

Proven on our own work

Nine kinds of work we have used it for.

Every line is work we ran on it ourselves, between July 5 and September 12, 2026, across 83 workspaces. It is not a menu of features, and none of it ships as a template; it is a record of what we happened to need. Bring it something that is not on this list and it staffs that instead.

Written deliverables · 30

Memos, briefs, investor content, sole-source justifications, requirement analyses, recurring recaps and debriefs, and the slide decks that carry them. A researcher, a domain analyst, a writer, a reviewer, and a scribe that renders to your own template.

Research & intelligence · 29

Competitor sweeps, market and segment scans, domain discovery, stakeholder mapping, use-case discovery, outreach and contact lists. Collectors gather, an analyst reconciles, a reviewer checks every claim back to a source.

Applications & pages · 20

Dashboards, trackers, session hubs and showcase apps, built, converted between formats, and repaired when something upstream moved. A builder, a designer, and a reviewer who opens the thing before you do.

Brand & marketing systems · 18

Whole brand systems, wordmarks, editorial blueprints, campaign and buyer narrative, content architecture and copy, image direction. A strategist sets the frame, a designer and a writer work to it, a reviewer holds the rules.

Code & platform audit · 18

Source-code and platform audits, codebase discovery, documentation audits, capability validation, diagnosis of somebody else’s system. Readers map the surface, specialists go deep on their area, a lead assembles the finding list.

Specialist creation · 16

Standing up a new expert for a named domain, a leadership roster, or a whole team in a fresh workspace. The foundry, used on its own. The domains we briefed it on range from enterprise sales tooling to professional sport, deliberately, to find the edge of it.

Financial & market work · 15

Filing retrieval and financial discovery, live market data pulls, position review, desk pipelines, connectivity probes. Data specialists pull, an analyst reconciles, a risk reviewer argues the other side.

Security testing · 10

Threat modelling, attack-surface mapping, engagement scoping, re-verification waves, jailbreak and compliance tests. Forged hunters work the surface; one document per finding, with file, line, path, fix and test.

Data extraction · 7

Pulling structure out of documents, screenshots, articles and feeds until it is something you can query. An extractor per source type, then a reconciler that makes them agree.

Kinds of work we ran

Workspaces we created

Plans the system wrote

177

Documents & pages it produced

796

Our own use, counted across both of our nodes, July 5 – September 12, 2026. None of these plans or specialists ship with the product.

Three ways to start

Use it. Integrate it. Roll it out.

Use it

You are the operator.

A workspace, a goal in plain language, a plan you read and sign. By Friday the workspace is worth more than it was on Monday.

Starts with · a goal

Integrate it

You are the builder.

Your product calls the API with a scoped machine key, a per-tenant ceiling and a scope check on every call.

Starts with · a scoped key and a ceiling

Roll it out

Your security team signs off.

A workspace per team on your own servers, your recurring workflows written as signed plans, gates where your policy says they go.

Starts with · the posture page

Or decide whether to build it yourself →

Next step

Thirty minutes, a real workspace, your own goal.

Bring something that is not on the list above. That is the better test.

Write to the assistant inbox

assistant@agenticensemble.ai

Request early access