How a session in Washington DC is put together
Every keynote is tailored after a briefing call, and the larger engagements start before the day rather than on it: a diagnostic of the client's own people, so the room hears its own evidence rather than somebody else's. What the room leaves with is written down.
What that has looked like: an agency business convening Asia, Asia-Pacific, EMEA and North America into one event, where the work was naming the recurring situations nobody had vocabulary for, and the company took that wording into its strategy; and a division of a very large company permitted exactly one AI tool, which after the constraint was examined moved to a multi-tool sandbox with different functions choosing different models, then repeated across further divisions.
Six of the seven engagements published on this site produced further work after the session rather than at it. The seven are set out in full at case studies, each with a line saying what it does not show.
How to read these examples. They are published to show how the work is put together, not as a client list for Washington DC. Most engagements are covered by confidentiality and are described without naming anyone; where a client is named anywhere on this site, it is with their agreement.
Two memoranda, thirteen months apart
On 28 March 2024, OMB memorandum M-24-10 set the rules for federal agency use of AI. It used the term automation bias twice. Once as a formal definition: the propensity for humans to inordinately favour suggestions from automated decision-making systems and to ignore or fail to seek out contradictory information made without automation. And once as a binding minimum practice, at section 5(c)(iv)(G), requiring agencies to ensure sufficient training, assessment and oversight for operators to combat any human-machine teaming issues (such as automation bias).
On 3 April 2025, M-25-21 rescinded and replaced it. The successor still requires human oversight, intervention and accountability for high-impact uses, and a route to timely human review and appeal. Read the whole document and the phrase automation bias appears nowhere in it.
So the obligation survived and the name of the thing it guards against did not. That is the most precise illustration this research has of its own argument: the control stays in the document, the reason for the control quietly leaves. Both memoranda were read in full, and the honest limit is recorded with the finding: over-reliance, deskilling and complacency appear in neither, so this is one term rather than a vocabulary.
The categories changed too. M-24-10 distinguished rights-impacting from safety-impacting AI, each with its own definition and its own presumptive list. M-25-21 collapses both into a single high-impact class. The mechanism is at what is automation bias.
The number a federal audience already owns
The 2025 federal AI use case inventory records 3,611 individually reported AI use cases across government, of which 445 are classed high-impact. Two hundred and fifteen of those sit in the Department of Veterans Affairs alone.
That is not a warning. It is an operating reality, self-reported, and it is the right starting point for a room that has spent two years being told what AI might do. The question a federal leadership audience can act on is narrower: of those high-impact cases, how many have a named person who could reconstruct the decision without the system, and how would you find out?
One gap stated plainly. No United States federal statistical agency measures what AI is doing to entry-level hiring. Both the Federal Reserve Bank of New York and Statistics Canada point outward to academic work for that, and so does this keynote. The evidence is at missing rungs, graded for what it does and does not prove.
One fragility worth naming from the stage
M-25-21 is a memorandum, not a statute. It was issued by a director and it can be replaced by the next one, which is exactly what happened to its predecessor. An organisation that has built its oversight arrangements to satisfy a memorandum has built them on something that changed once inside two years.
Which is why the useful version of this session for a Washington audience is not about compliance. It is about what an agency would still be able to do if the guidance changed again: who can check the work, and whether they could do the work. That question is at who supervises work they cannot do.
Who books this in Washington, and who should not
Federal leadership and senior executive service audiences, agency chief AI officers and their teams, association and trade-body annual meetings, government contractors and the consultancies around them, and think-tank and policy convenings.
He is the wrong choice for a briefing on what the law should say, which is properly the ground of the people who write it. This session is the operational question that follows: given the obligation exists, what does an organisation have to be able to do, and how would it know it could?
The three keynotes a Washington DC room can book
Keynotes run 30 to 90 minutes, in person or virtual. A 30-minute main-stage version of each talk is available for conferences; the full argument needs 40 or more. The sector examples change for the room; the argument does not. All three, with the showreel, are at keynotes, and the programme copy is at the speaker pack.
Drift versus Design
Drift versus Design: why most organisations hand their judgement to AI without deciding to, and how to design your way through instead.
Most organisations are adopting AI by drift: a thousand reasonable decisions that add up to judgement nobody chose to give away. Rahim Hirji, author of SuperSkills, shows leaders which mode they are running, where AI sharpens judgement and where it weakens it, and hands them the controls.
We Are Superheroes
We Are Superheroes: AI is the suit. The human decides.
AI makes everyone faster, stronger and more capable. It does not decide; people do. Through a family story across four generations and three continents, Rahim Hirji hands the audience the seven human skills that grow more valuable as the tools spread. They arrive thinking AI is the story and leave knowing they are.
WTH (What the Human)
WTH (What the Human): a live test of a board's own judgement.
Business is being rewired as AI arrives, and human judgement is leaving with it. In three acts, Rahim Hirji shows a board what AI is changing in how organisations run, then ends with a live test of the board's own judgement, in the room and in real time. Nobody forgets the result.
Where these sessions happen in Washington
A shorter list than most cities on this site, because Washington concentrates its large conference capacity in very few buildings and does most of its serious convening in rooms that publish nothing.
- Walter E. Washington Convention Center, Mount Vernon Square
- 2.3 million square feet, with 703,000 of exhibit space, and the venue states it takes events from 500 to 42,000 attendees. Metro-connected. The only building in the District at association-annual-meeting scale.
- Ronald Reagan Building and International Trade Center, Pennsylvania Avenue
- The federal government's own conference venue, on the avenue between the White House and the Capitol. The address does work that no ballroom does when the audience is federal.
- The K Street corridor and downtown hotels
- Where most association and contractor business actually happens: hotel ballrooms rather than purpose-built conference floors, which means flat floors, risers and screens. Worth settling the room shape before the session is designed rather than after.
One scheduling note specific to this city and it is not seasonal. The federal fiscal year ends on 30 September, and the weeks either side of it are the hardest in the calendar for a federal or contractor audience. Congressional recesses move the policy-adjacent audience in and out of town on a published schedule, so a session aimed at Hill-facing staff should be checked against it rather than against the weather.
