Capability moved faster than intervention.
AI systems are becoming more capable. Organisations are becoming more dependent on automation. Human oversight only matters if intervention remains possible.
Intervene's methodology is developed in academic collaboration with Prof. Nicholas Ryman-Tubb, University of Surrey, calibrated against a hold-out corpus of regulated automation and intervention failure modes.
Founder · Principal Advisor
Most organisations can evidence governance. Far fewer can evidence intervention capability.
For years I watched organisations invest heavily in governance, controls, reporting and assurance. As AI systems became more capable and organisations became more dependent on automation, the same pattern kept becoming more consequential. The controls existed. The reporting existed. The governance structure existed.
The intervention did not.
Problems were detected too late. Escalations stalled. Decisions were delayed. Ownership became unclear. By the time action was taken, the opportunity to prevent harm had already passed.
That gap became the foundation of AGDA™.
My background spans more than three decades across technology, systems architecture, digital products, payments, operational systems and organisational risk. I was also a co-founder of the Institute of Financial Innovation in Transactions and Security (IFTS), an industry initiative focused on improving trust, security and resilience in financial systems.
Throughout my career I have been drawn to the same challenge: understanding how complex systems behave under pressure, and why organisations often struggle to act when it matters most.
The emergence of AI brought that challenge into sharper focus: systems can act faster, affect more people, and create consequences that become harder to reverse.
Many organisations can describe their governance structures. Far fewer can demonstrate that they are capable of detecting, escalating, deciding and intervening before consequences become irreversible. Regulators increasingly require effective human oversight. Boards increasingly need confidence that intervention remains possible.
AGDA™ was developed to answer that question directly.
Rather than assessing whether governance controls exist on paper, AGDA™ measures Intervention Readiness: whether intervention is actually possible within the time available. The methodology produces a deterministic, independently verifiable verdict supported by cryptographic attestation and board-ready reporting.
Today, through Intervene, I work with organisations that need a clear answer to a difficult question:
Can you stop it in time?
Because when the answer matters most, assumptions are not enough.
The operating commitments behind the instrument.
Operating constraints, codified into how we accept work, how we calibrate, what we will and will not deliver.
-
Independent judgement.
No vendor affiliation, no referral fees, no co-marketing. We do not implement what we recommend.
-
Published methodology.
Theses, catalogues, and rater protocols are public and citeable. Calibration held private as trade secret.
-
Verifiable computation.
Every verdict Ed25519 signed. Inputs and outputs hashed. An audit committee can verify without contacting us.
-
Refusal of false control.
If the evidence does not support green, the engine will not produce green. Findings the client did not want, in writing, before the regulator does.
Where the category is most urgent.
Three primary sectors. Two by invitation. Some we deliberately do not serve.
-
UK-regulated financial services.
Banks, asset managers, insurers under PRA / FCA jurisdiction with material automated decision systems.
-
EU high-risk AI operators.
Annex III high-risk systems under the EU AI Act, where human oversight must be capable of intervention.
-
Critical national infrastructure.
Energy, water, telecoms, transport. Operators where intervention failure compounds across dependency chains.
-
Healthcare systems.
Clinical decision support and diagnostic AI. By invitation, where independence from device vendors can be guaranteed.
-
Defence & intelligence.
Autonomous decision systems and human-on-the-loop frameworks. With appropriate clearances and engagement letters.
Consumer internet, enterprise productivity, B2B SaaS, advertising tech. Not because the work is unimportant, but because Intervene's calibration corpus and rater pool are tuned to regulated, board-accountable systems.
What we are not.
About the practice, not the product. (For product-category boundaries, see What AGDA™ is not on Services.) Good practices occupy the adjacent ground. Knowing the difference avoids procurement confusion later.
-
Not a compliance dashboard.
Compliance tools handle evidence management. We measure whether intervention capability holds under stress.
-
Not a Big-4 advisory.
No leveraged teams, no slide decks, no implementation arm. The output is a signed verdict.
-
Not a policy-writing practice.
We do not draft your automation policy. We measure whether the controls behind it produce what the policy claims.
-
Not a model-testing lab.
Red-team labs test models. We test whether the chain around the model can act on what the lab finds.
-
Not an implementation partner.
No standing-up of controls, no platform integration, no runbook ops. Remediation belongs to you or someone else.
-
Not a safety research practice.
Operational, not philosophical. A verdict on a specific system, not a paper on risk in general.
The question returns to the same place.
Can you stop it in time? See the evidence format before opening a commercial conversation.