Agent Centric AI
Agent Centric AI · Commercial synopsis Prognos Agentic Engine

Give your best people
an Iron Man suit .
Not a replacement.

The commercial power of Agent Centric AI is not avoiding new hires — it is equipping the people you already trust with an operational exoskeleton that multiplies their daily throughput. The Prognos Agentic Engine deploys agents that sit under your senior practitioners, move the work, and hand every judgment call back.

14.8s
Site photo
to engineer action
<10 min
Basel 3 CAR review
on live workbooks
90 min
Meeting recording
to instructing brief
0
Judgment calls
delegated to an agent
Live · Prognos /site-ops · Edinburgh 04:21 UTC
Naledi · Site-Ops Agent
run_id 7f2a · project BRAEMAR-04C
ELAPSED
00:14.8
Inbound
Photo from foreman (+27 82 644 1083): column C3, Level 4 pour — cross-check against drawing DWG-04C-rev08|
  1. Ingested site photo · EXIF 04:19:42 UTC 1.2s
  2. Cross-referenced DWG-04C-rev08 · 214 anchors matched 4.6s
  3. Deviation flagged: rebar spacing at C3 off by 47mm 6.1s
  4. Opened action ACT-3118 · owner D. Volkhov 2.1s
  5. Notifying engineer & PM Imran Rashid-Khan 0.8s
Judgment call → D. Volkhov, P.Eng
≤ defined boundary policy v3.1
One site photo in. One engineer action out. No project manager chased it. Prognos agents move the work; people own the outcome.
Prognos /core v2.4 Basel III · CAR module IFRS 17 reconciliation Counterparty memory · 11.4M records Signed by Humans only Site-Ops · Legal-Ops · Risk-Ops Edinburgh · Johannesburg · London
§ 01 · The hidden tax

Most teams are paying a tax
nobody writes down.

Coordination debt, management drag, and senior attention spent on low-grade operational friction. Traditional software only records this work. Prognos-powered agents actively move it — so your most expensive minds stop doing the cheapest work in the building.

Observed across 17 deployments, 2024–26
Chasing updates & reconciling status
31.8%
· agent
Formatting & assembling documents
18.4%
· agent
Retrieving data across systems
14.6%
· agent
Multi-party scheduling & rescheduling
9.1%
· agent
Judgment, strategy & relationships
26.1%
· human
With a Prognos deployment, the first four rows are absorbed by specialist agents. The fifth row — the one that actually pays the salary — is the only row your people should be spending time on.
§ 02 · The suit, in action

What 10× productivity
actually looks like.

Four live deployments, reconstructed end-to-end. Every decision point in a coloured band is returned to a human.

Site-Ops / BRAEMAR-04C
15.0s end-to-end
Site photo — column C3, Level 4
C3 · 47mm off
IMG_4418 · 04:19:42 UTC
C3 DWG-04C-rev08
214 anchors matched
ACT-3118 Priority · P1
Re-dress rebar spacing at C3 (Level 4) — 47mm deviation from DWG-04C-rev08.
DV D. Volkhov, P.Eng
ETA 06:00
Agent move Deviation flagged Human decision
No PM touched a keyboard.
Site inspection, end to end, in 15 seconds.
Ingest the foreman's photo, cross-reference architectural drawings, flag the deviation, open an action, and notify the engineer — autonomously, inside defined boundaries.
Coord / scheduling
4 parties · 3 TZ
NM
Naledi Makhubu · Edinburgh
free 14:30–16:00
+0h
JE
Johanna Engelhardt · Frankfurt
conflict 15:00 · moved
+1h
IR
Imran Rashid-Khan · Karachi
confirmed
+4h
DF
Dunbarton & Finch (legal)
confirmed
Tuesday, 03 March · 14:30 UTC
No keyboard touched
Multi-party conflicts, resolved silently.
Four parties, three time zones, one meeting. Your assistant never saw it.
Risk-Ops / Basel III
09:47 elapsed
CET1 ratio
14.37%
+0.41 pp wk
Total CAR
17.82%
-0.12 pp wk
RWA drift
+R 1.84bn
corp book
Lever surfaced
Restructure exposure to Stenbeck-Westhaus facility
Δ CAR +38bp
Running stress scenario BCBS-S08b on live workbook CAR-Q1-26.xlsx
4 of 7 · 02:11 remaining
Hands back to Dmitri Volkhov, CRO < 10 min target
Basel 3 CAR review, under 10 minutes.
Complex regulatory checks against live workbooks, surfaced as three real levers a human can pull. Not a dashboard — a decision.
Legal-Ops / Dunbarton & Finch
87:14 elapsed · ≤ 90 min target
Instructing brief · draft 3
Re: Proposed facility amendment — Stenbeck-Westhaus Capital (counterparty ref SWC-4718).
Counsel is instructed to draft a side-letter amending clauses 7.2 and 12.4 of the 2023 facility, reflecting the revised covenant tested in the 12 Feb steering call (recording ref MTG-8831, 00:41:18–00:47:52).
Historical dealings: four prior amendments on this counterparty, each with an escalating materiality threshold. The team proposes matching the 2024 pattern…|
Sources cross-referenced
  • 01 MTG-8831 · steering call transcript
  • 02 Facility agreement, 2023 executed
  • 03 4 × prior SWC amendments
  • 04 Counterparty memory · 11.4M rows
Returned to human
JE Johanna Engelhardt · sign or send
3 clauses flagged for review
Draft assembled from meeting recording · cross-referenced against historical counterparty data v3 · 2h 41m ahead of SLA
From meeting recording to instructing brief, in 90 minutes.
Instead of half a day of drafting, your lawyer opens a clean v3 with every historical clause already reconciled. They still sign.
Live fleet · last 24h
21 tenants · 47.2% median throughput lift
14.8ssite photo → action 9:47Basel CAR → lever 87:14brief draft 3 · SWC-4718 00:03counterparty diff · 11.4M rows 47.2%median throughput lift +1 (312) 847-1928ops hotline 0judgment calls delegated
Live telemetry from the production fleet. Nothing cached, nothing rounded.
§ 03 · Economics

More output.
Same headcount.

An agentic ecosystem absorbs the repeatable operational volume without the delay, inconsistency, and supervision burden of scaling junior teams. Your existing people gain perfect institutional memory, live-state awareness of the business, and the ability to execute across dozens of channels at once.

47.2%
Median throughput lift across 17 deployed teams.
3.5d → 90m
Average collapse of legal drafting cycle time.
11.4M
Counterparty interactions retained in institutional memory.
0
Judgment calls transferred to an agent. Ever.
Never sleeps. Agents run continuously inside defined boundaries, 24/7.
Never loses context. Every email, meeting and filing stays addressable forever.
Acts across dozens of channels simultaneously. One team, full-institution coverage.
§ 04 · Operating model

Humans at the centre.
Always.

Prognos does not cut humans out of the loop or outsource judgment. It places your practitioners at the centre and hands the final decision back, every time. The technology does not replace roles — it replaces the friction, delay and administrative drag between them.

“We don't replace humans. We replace the friction between them.”
Agent Centric AI · design charter §1.2
What Prognos does
Moves coordination, retrieval and formatting work continuously, within policy.
What Prognos refuses
Signing, deciding, approving, or negotiating on behalf of a human.
Operates within defined boundaries
Every agent has a policy envelope: systems it can touch, actions it can take, thresholds that escalate.
Never acts past the boundary
At the edge of policy, the agent stops and hands back to a named human, not a queue.
Holds perfect institutional memory
Every interaction, filing and meeting stays addressable — so context never walks out of the door with a resignation.
No black-box automation
Every agent move is logged with sources, reasoning, and the named human who owned it.
§ 05 · Deployments

Where the suit is already
on the floor.

A selection of live Prognos Agentic Engine deployments. Names redacted where required by counterparty.

01 · Civil eng.
Braemar Civil Engineering
Edinburgh · 340 engineers · 11 live sites
Site-Ops agent absorbed photo-to-action triage across all 11 active sites. Median inspection-to-ticket time collapsed from 41 min to 14.8 s. PMs no longer chase updates.
02 · Banking
Stenbeck & Westhaus Group
Stockholm / Johannesburg · Tier 2 wholesale
Risk-Ops agent reviews Basel III CAR against live workbooks nightly, surfaces three actionable levers for the CRO by 06:00. Saves an estimated 4.2 FTE of senior analyst time per month.
03 · Legal
Dunbarton & Finch LLP
London · Banking & projects practice
Legal-Ops agent converts steering-call recordings into draft instructing briefs in 90 min, cross-referenced against 11.4M counterparty rows. Partners review, not re-type.
04 · Asset mgmt.
Redacted · European asset manager
€38bn AUM · private markets book
Coord agent silently resolves multi-party scheduling across 3 time zones and 6 counterparties per deal. 0 assistants touched a keyboard last quarter.
§ 06 · Pair your team with the Prognos Agentic Engine

Hand your team the suit.
Keep the pilot seat.

A 6-week pilot deploys one specialist agent onto a single team, inside your security boundary. No new headcount. No retraining. Your people continue to own every decision — they just stop doing the cheapest work in the building.

Request a pilot deployment Read the operating model brief
pilots@agentcentric.ai +1 (312) 847-1928