Advisory · Agentic AI
Speed − Trust = Liability.
Agentic Transformationthat lasts.
Independent readiness. Phased build. Annual durability. For organizations crossing from assistive AI to autonomous agents.
Anchored to the A7 framework. Not a notified body. Not a platform reseller. Not for sale on volume.
The clock that matters
The honest question is not whether your programme is moving fast enough.
It is whether it is ready at all.
Regulator
2 Dec 2027
EU AI Act Article 6 high-risk obligations apply to standalone Annex III systems — deferred from Aug 2026 by the Digital Omnibus (provisional agreement 7 May 2026, pending Official Journal publication).
Annex III triggers in financial services include creditworthiness scoring (§5(b)). For most banks, insurers, and asset managers operating in EU jurisdictions, the obligations land here.
↳ EU AI Act, Art. 6 + Annex III · Digital Omnibus provisional agreement, 7 May 2026
Incident · 23 Jul 2025
Replit
An AI agent deleted a production database. The missing engineering control: dual-control on production-write actions.
The agent had unbounded tool-use access — no allow-list, no kill-switch, no second factor on destructive operations. This is the Tier-1 Technical Readiness Floor failure case.
↳ Fortune, 23 July 2025
Incident · Dec 2025
Amazon Kiro
An autonomous AI agent reportedly deleted and recreated a production environment in a regional deployment.
Same architectural property as Replit: an L3-equivalent autonomy deployed without dual-control. Reported by industry analysts, not regulator filings — flagged as such.
↳ Industry reporting, December 2025
Market · 25 Jun 2025
> 40 %
of agentic AI projects will be cancelled by end-2027 — Gartner forecast.
~40% of engagements being sold today should not have been started. Procurement-mandated independent advisors exist precisely to filter the other 40%.
↳ Gartner press release, 25 June 2025
Hover any card for the source · all dated, all primary
Three services
Three services. Each earns the next.
While the Big-4 sell 1,000-agent libraries, this practice ships one verdict at a time.
A7 Agentic Readiness Assessment
A focused engagement
An independent verdict on whether your organization is ready for autonomy — scored against the seven A7 dimensions. "Not yet" is a first-class outcome.
What you receive
- 01A7 score-card across seven dimensions
- 02Regulatory exposure mapped to your use cases
- 03Liability & decision-rights gap analysis
- 04Verdict letter, signed personally
- 0590-minute ExCo readout
Phased Autonomy Build
Following a clearing verdict
Engaged only after the assessment clears. Delivered to L1 or L2 only. Governance scaffolding built to survive the next audit.
What you receive
- 01Pilot scoped with named stop-conditions
- 02Production-grade Agent Register & Oversight Log
- 03Validated kill-switch under load
- 04Audit trail tested against regulator question
- 05GC sign-off on liability framework as deployed
Annual Durability Review
Once a year, going forward
The proof that the transformation is still working. Allowed to fail — "regression detected" is a real outcome, named honestly.
What you receive
- 01A7 re-scored against current state
- 02Regulatory delta — what changed since last review
- 03Every Oversight Log entry walked through
- 04RACI audit — has the framework drifted
- 05One-page Durability Verdict
The brand commitment
I will tell youno.
Big 4 firms cannot afford to tell you that you are not ready — their utilization targets reward delivery, not refusal. I can.
The “not yet” verdict is not a marketing line. It is a version of the assessment that returns a remediation roadmap instead of a build plan, and it is the most important version this engagement produces.
If you need permission to proceed, this is the wrong engagement.
What triggers a “not yet”
Three absolute floors. Failure of any one returns “not yet.”
- —Unbounded tool-use — no allow-list, no kill-switch
- —No agent observability — no per-step trace, no replay
- —No evaluation harness for the agent’s action space
- —No named accountable executive for autonomous decisions
- —No documented decision-rights matrix
- —No incident-response runbook with <60-min escalation
- —No EU AI Act Annex III mapping per use case
- —No SR 11-7 / SS1/23 model-risk integration in FS
- —No NIST AI 600-1 incident disclosure protocol
The framework
Seven dimensions. Five autonomy levels. One verdict.
A7 scores each dimension against the autonomy level your organization is targeting. Click any cell.
A2 — Technical Infrastructure
↳ What passes at this level
Standardized. Containerized (Docker/K8s), basic CI/CD. Agents run as standalone processes. No undo capability.
AI acts within bounded action space. Human reviews each batch.
The white space
Every major firm sells the same three things. None sells the fourth.
(i) a branded framework, (ii) a launched platform partnership, (iii) an industry agent library. None publicly leads with refusal — a first-class “not yet.”
AskAjay.ai
Independent. Verdict-first. "Not yet" is a real outcome.
Hover or tap any firm on the map
The federation
Where the assessment surfaces an operational gap, the open-source tools at AiExponent.com are the layer your team can adopt.
RiskForge for EU AI Act mapping. license-compliance-checker for stack hygiene. RAG-Bench for evaluation rigour. They are referenced, not bundled. This service is independent of any vendor — including the AiExponent product line.
Begin a conversation
If autonomy is on your roadmap, the conversation starts here.
Engagements are taken on a small number per year, by direct conversation only. No volume pipeline, no associate substitution. The first call is the founder.
How conversations begin
- 01
You write to me directly
Role, organization, geography, current AI maturity, and the trigger event behind your interest.
- 02
45-min founder call
Under non-marketing framing. We confirm fit on both sides.
- 03
Engagement letter or referral
A "not in scope" outcome is a real possibility — and is named honestly at this stage.