Digital workforce
A workforce has a shift pattern, a manager and a disciplinary record. A chatbot has none of those.
Every co-worker below is scoped to one job, bound to the systems that job needs and nothing else, evaluated continuously against a suite it cannot edit, and required to stop at the points where policy says a person decides. Squads scale by cloning a proven role, not by writing a new prompt. Adding the CIT to the estate added eleven co-workers in a morning.
Co-workers on shift
268
across 7 squads, 3 time zones
Actions / 24h
9,283
reads, drafts, checks, routings
Stopped at a control
49
policy said a person decides
Suspended by policy
4
fell below their evaluation suite
Human hours returned
1,940 / wk
measured against the task baseline
A.The floor — every co-worker on shift, by squad
One cell is one co-worker. Colour is its state right now, not its health score.
state sampled --:--:-- ET · refreshes every 1.5s
Launch orchestration
wrapper launches, dependencies, approvals
36
co-workers
612
actions / 24h
9
to a human
Product and distribution
eligibility, terms, approved materials
54
co-workers
1,884
actions / 24h
7
to a human
Credit data and valuation
loan tape, pricing feeds, marks, lineage
64
co-workers
2,431
actions / 24h
14
to a human
Retirement and fiduciary
DDQs, sponsor packs, committee material
39
co-workers
408
actions / 24h
11
to a human
Partner and market signal
adviser, consultant and platform feedback
31
co-workers
996
actions / 24h
3
to a human
Investor reporting
statements, factsheets, regulatory returns
27
co-workers
744
actions / 24h
5
to a human
Control and assurance
evidence, audit trail, evaluation of the rest
17
co-workers
2,208
actions / 24h
0
to a human
Squads are org units, not models. A co-worker joins one by being given a role, a scope and an evaluation suite — the same three things a new analyst is given.
Busiest co-workers, last 24 hours
Sorted by actions taken. Success is measured against the co-worker's own evaluation suite, not a rating.
| Co-worker | Squad | What it does | Actions | Passing | Escalated |
|---|---|---|---|---|---|
eltif-eligibility-check | Product and distribution | answers can this client, in this country, hold this share class | 1,204 | 99.1% | 2 |
daily-mark-reconciler | Credit data and valuation | reconciles the daily estimate against three independent inputs | 968 | 97.4% | 6 |
loan-tape-normaliser | Credit data and valuation | maps each originator file into the standard loan schema | 884 | 94.2% | 5 |
material-clearance | Product and distribution | checks every deck and factsheet against the approved language set | 641 | 98.8% | 3 |
ddq-drafter | Retirement and fiduciary | drafts questionnaire answers from the fund record and prior filings | 318 | 92.6% | 11 |
cit-unit-value-builder | Credit data and valuation | builds the recordkeeper unit-value file for the US DC trust | 246 | 96.0% | 4 |
What stops this becoming shadow IT
DAF is the regulator of the workforce. These are its five standing controls.
Scoped access, not blanket access
19 connectorsA co-worker reaches the systems its role needs. The eligibility checker cannot read the loan tape.
Continuous evaluation
31 suitesEach role carries a suite it cannot edit. Falling below it suspends the co-worker automatically.
Mandatory stopping points
49 todayMarks, disclosures, client-facing language and anything sent to a regulator stop for a named person.
Evidence by default
100% of actionsInput, source, model, output and approver are recorded per action and are queryable as one file.
A named human owner
7 squadsEvery squad has a manager in the business, not in engineering. Performance is reviewed like a team's.