Diffusion is now live. Read more

05 / Speaking

Book the leading voice in software factory implementation.

Justin McCarthy takes the frontier patterns behind reliable AI software work to stages, studios, and panels — for leaders moving from agent demos to systems their institutions can trust.

Founder, Diffusion · former Co-Founder & CTO, StrongDM

Invite Justin to Speak
Justin McCarthy speaking on stage in front of an adoption-curve chart

Starting is solved. Finishing is the whole job.

AI can start almost anything now — a demo, a migration, a first draft. The frontier is the last mile: intent that survives translation, acceptance tests the model can't argue with, permissions that hold at runtime, and teams that supervise more agents without drowning in attention debt. Closing that last mile — into systems institutions can trust — is where the real work now lives.

Diffusion turns the patterns advanced AI builders are discovering into an operating model for complex organizations: attractor pipelines that return to a known-good state, critic and validator apps that act as law, digital twins for unsafe failure rehearsal, and decision-theater tools that give stakeholders a concrete future to react to.

Leave with a map of the last mile

Where demos stall, what finishing actually costs, and how to tell whether a pipeline is stabilizing or just producing motion.

Last mile map routing
SEC 01 · ORG/EDGE SEC 04 · PRODUCTION X0 Y720 X1200 Y0 HAPPY-PATH OPTIMISM ✕ DEAD END INTENT DRIFT ✕ DEAD END MISSING ACCEPTANCE TESTS ✕ DEAD END UNSAFE PERMISSIONS ✕ DEAD END NO RUNTIME OBSERVABILITY ✕ DEAD END DEMO INTENT TESTS PERMISSIONS RUNTIME TRUSTED SYSTEM

The route

One valid pathway to a trusted system — and five dead ends. Drag the map; hover any node.

A pattern set: critics, twins, attractors

Concrete patterns for validation, rehearsal, permissions, and runtime control — explained for builders and decision-makers in the same room.

cycling
PASS PASS PASS PASS FAIL FAIL FAIL FAIL ATTRACTOR PULL RESTART FROM KNOWN-GOOD AGENT CRITIC VALIDATOR DIGITAL TWIN RUNTIME LAW KNOWN-GOOD STATE

The loop

Six parts, one loop. Failed work never ships — the attractor pulls it back to the last known-good state.

An org model: attention as infrastructure

Why attention is the scarce resource — how many agents one person, team, or institution can responsibly supervise.

Attention load map stable
1 TEAM RESPONSIBLE SUPERVISION
Agents
Autonomy
Blast radius

5 agents · assisted · low blast radius

Clear ownership, low exception load.

Signature talks

Five understandings,
shaped to the room.

Each runs as a keynote, a panel seat, or a long-form conversation. Titles are starting points — the argument gets tuned to your room.

  1. T-01

    The Software Factory

    Starting is solved, but finishing is a moving target. Understand what it takes to carry AI-generated engineering to production-grade reliability. This means workflows and pipelines built around verifiability, veracity, and observability.

    Keynote · Podcast
  2. T-02

    Decision Theater

    Most stakeholders are better editors than authors. In this workshop, we build cheap, disposable software in realtime to demonstrate concrete futures people can react to, so the organization can extract intent before it commits.

    Keynote · Workshop
  3. T-03

    Critics, Twins & Runtime Law

    Agentic software governance as first-class design: we explore how digital twins act as validation enforcement, rehearsing and obviating failures you can't trigger live.

    Panel · Podcast
  4. T-04

    When the Rules Invert

    Yesterday's engineering truths are today's engineering albatross: Rewrite from scratch. Spend as much on tools as on the product. Re-platform on a whim. Why the economics flipped — and which hard-won instincts to unlearn.

    Keynote · Fireside
  5. T-05

    The Organization After Agents

    Taste, attention, and span of control are the true bottlenecks in your team: understand how many agents one person can realistically supervise, where hierarchy helps, and what careers become when people run sections of agents.

    Keynote · Workshop

Formats

In person or remote.
Room-sized to hall-sized.

Justin McCarthy taking questions during a live talk

Keynote

20–45 min. Single argument, built for a decision-making audience.

Conference panel

Seat or moderator. Strong with contrarian or technical lineups.

Podcast

Long-form, remote or in-studio. Technical depth welcome.

Fireside chat

Interviewed format for summits and invite-only rooms.

Working session

Lecture or workshop for leadership teams and cohorts.

Invite Justin to speak.

Fill us in on the event, format, audience, and window. We'll get back to you quickly with a real answer from a real person.

Invite Justin to Speak