How to Book the Right AI Speaker for Your Event or Conference

How to Find and Book a Good AI Speaker: The Short Answer

Skip the celebrity circuit and find someone who has shipped an AI system to production in the last 18 months, can name the eval framework they used, and will not melt under hostile technical questions from your engineers. That single filter eliminates roughly 80 percent of the people currently marketing themselves as AI speakers.

I am Mahmoud Zalt, an independent senior AI systems architect with 16+ years building production software since 2010. I founded Sista AI, where the past year has been spent running a live workforce of autonomous agents in production and deliver AI workshops, training sessions, and conference talks for engineering teams and technical leadership. You can read more about my work on the about page or browse projects. This article gives you the exact process I would use to vet a speaker if I were booking one for my own event.

Why Most AI Speaker Searches Go Wrong

The AI speaker market is flooded right now. LinkedIn shows roughly three categories:

• The pundit. Frequent conference appearances, polished decks, general 'AI is transforming everything' narratives. Has not written production code in years. Sounds impressive until your senior engineers start asking follow-up questions.
• The vendor rep. Employed by an AI company. Talk is a product pitch wearing a conference badge. Fine if your audience wants demos, dangerous if they want objective guidance.
• The practitioner. Has shipped systems, burned money on failed experiments, knows what guardrails actually look like in production, and can say 'I don't know, but here's how I would find out.' Harder to find, worth the extra effort.

The problem is that most RFPs and speaker booking processes are optimized for name recognition, not practitioner depth. You end up paying more for less.

The Six-Question Vetting Checklist

Before you reach out to any speaker, run them through these six questions. A practitioner answers them specifically. A pundit either avoids them or gives generic answers.

1. What did you ship in the last 18 months? You want: a specific system, team size, stack, and outcome. Red flag: 'I advised many organizations on AI strategy.'
2. What went wrong and how did you catch it? Production AI fails in specific ways: retrieval drift, prompt injection, tool-call loops, eval regression after a model version bump. Anyone who has actually shipped knows at least two war stories. Red flag: 'Our projects generally went smoothly.'
3. How do you evaluate your models? You want a named methodology: RAGAS, G-Eval, LLM-as-judge with an adversarial probe set, human preference labeling cadence. Red flag: 'We look at accuracy metrics.'
4. Can you field live technical questions from senior engineers? Ask them to name a topic your engineers care about and invite them to describe what a hostile question looks like and how they'd handle it. Red flag: hesitation or 'I like to keep talks accessible to everyone.'
5. What is your guardrails and safety architecture? Good answer names specific layers: input validation, output filtering, tool-call sandboxing, human-in-the-loop checkpoints, rate limiting, audit logs. Red flag: 'We use responsible AI principles.'
6. What would you recommend we NOT build right now? The ability to tell a room of engineers to slow down on a specific pattern, and why, is the clearest signal of real judgment. Pundits rarely say no to anything.

Match the Format to Your Audience

The right speaker depends heavily on the format you are running. Here is a quick mapping:

The biggest mistake I see: booking a keynote-style pundit for a room of senior engineers. Engineers will disengage within 15 minutes when they realize the speaker cannot answer a specific question about context window management or vector store indexing strategies. Flip the logic: optimize for the hardest questions in the room, not the easiest.

Skip the Celebrity Headliner

Here is the contrarian advice: a recognizable name rarely delivers proportional value for technical audiences. The $30k keynote from a recognizable AI figure often produces a generic talk that could have been a blog post. The $8k practitioner who ran a production agentic pipeline for six months will generate better post-event feedback scores, more actionable takeaways, and conversations in the hallway that your engineers actually remember.

The celebrity headliner is useful for one specific job: drawing registrations to a consumer or executive conference where name recognition drives ticket sales. If that is your goal, book accordingly. If your goal is to level up your engineering team's judgment about real AI systems, you need depth, not fame.

A practical test: search the speaker's GitHub. Look for commits in the last year. Look for issue threads where they reason through a technical problem in public. Look for libraries other people use. That kind of signal is hard to fake and tells you more than any speaker bio.

A Worked Example: What Good Looks Like

Say you are running a two-day engineering summit for a company building a customer-facing AI assistant. You have 40 engineers, 8 tech leads, and 3 engineering directors. You want a half-day session on agentic architecture and a 45-minute talk the next morning on production readiness.

Here is what I would specify in the brief:

• Half-day workshop: cover agent loop design, tool-calling and MCP server patterns, retrieval pipeline with RAGAS evals, guardrails (input sanitization, output filtering, human-in-the-loop checkpoints), and a live exercise where teams redesign one of your existing workflows as an agentic system. Deliverable: a one-page architecture template each team takes back to their codebase.
• Morning talk: production AI observability. Specific topics: tracing LLM calls with structured logs, detecting retrieval drift, handling model version bumps without eval regression, cost attribution per feature. 40-minute talk, 20-minute open Q&A. No vendor endorsements.

When you brief a speaker this specifically and they push back with useful corrections ('your teams will get more value if we restructure the workshop to start with evals rather than architecture'), that is a practitioner. When they accept every requirement without question, that is someone optimizing for the booking, not the outcome.

Logistics, Lead Time, and Cost

Realistic cost ranges for a practitioner-level AI speaker in 2025 to 2026:

• Remote conference talk (40 to 60 min, includes prep and Q&A): $1,500 to $3,000
• On-site keynote (travel not included): $2,000 to $5,000
• Half-day interactive workshop: $4,000 to $7,000
• Full-day workshop with custom exercises and follow-up materials: $7,000 to $15,000
• Multi-day team training cohort: priced on scope, typically $15,000 and up

Lead time: for a single remote talk, 2 to 3 weeks is usually enough. For custom workshops, plan for 4 to 6 weeks minimum to allow proper scoping and exercise design. For large on-site conferences requiring travel, 8 to 12 weeks is safer.

What should be included by default: one scoping call, custom content alignment (not recycled slides), a slide deck or workshop materials shared with attendees after, and a post-event Q&A channel open for 2 weeks. If a speaker charges extra for any of these on a $10k engagement, that is a red flag.

What you should not pay for: a generic talk on 'the future of AI' with no connection to your audience's actual work. Always require a written brief confirming the specific topics and format before signing.

Frequently Asked Questions

how do I find a good AI speaker for a technical conference

Start with your network: ask engineering leaders you respect who they have seen present at meetups or internal events. Then check speaker lists from practitioner conferences like QCon, Strange Loop, or AI Engineer Summit. Look for speakers who submitted proposals and were accepted through a technical review process rather than speakers who were invited because of their company affiliation or social following. GitHub activity and public writing are stronger signals than a polished speaker page.

what is the difference between an AI keynote speaker and an AI workshop facilitator

A keynote is a one-to-many broadcast format: one person presents, the audience watches and absorbs. A workshop is interactive, hands-on, and requires the facilitator to respond dynamically to your specific team's questions and codebase. The skill sets overlap but are not identical. A great keynote speaker can be a mediocre workshop facilitator if they cannot improvise under live questions. Always ask specifically about the format you are booking and request a reference from someone who attended the same format, not just the same speaker.

how much does an AI conference speaker cost

Remote practitioner talks run $1,500 to $3,000. On-site keynotes add travel and typically land at $2,000 to $5,000 for a practitioner, significantly more for a celebrity name. Half-day workshops range from $4,000 to $7,000. Full-day engagements from $7,000 to $15,000. These are practitioner-level rates. Celebrity headliners often start at $20,000 to $50,000 and deliver proportionally less technical depth for engineering audiences.

should I book a well-known AI researcher or a practitioner for my engineering team

For an engineering team, almost always the practitioner. Researchers are excellent for academic conferences, graduate programs, or organizations specifically exploring cutting-edge research. For teams building production systems, a practitioner who has dealt with token cost management, production evals, RAG pipeline tuning, and agentic reliability will generate more actionable takeaways. The test is simple: ask the speaker to describe a specific failure they debugged in production. That answer tells you more than any credential.

what questions should I ask an AI speaker before booking

Six questions matter most: What did you ship recently and what was the outcome? What went wrong and how did you catch it? How do you evaluate AI systems? Can you handle hostile technical questions live? What does your guardrails architecture look like? And finally, what would you recommend we avoid building right now? A practitioner gives specific answers to all six. A pundit deflects, generalizes, or reframes them into talking points.

how far in advance should I book an AI speaker

For a single remote talk, 2 to 3 weeks is workable. For custom workshops with exercises designed around your team's stack, 4 to 6 weeks minimum. For large on-site conferences requiring travel coordination, 8 to 12 weeks. If the speaker you want is fully booked, ask about a waitlist or a virtual format: many practitioners who limit on-site travel have more availability for remote sessions.

Book Someone Who Can Answer Your Engineers' Hardest Questions

The right AI speaker for your event is the one your senior engineers will remember six months later, not because they were famous, but because they said something specific and true that changed how your team thinks about a real problem. That speaker exists. They are usually not the one with the biggest following or the highest speaking fee.

I deliver AI workshops, conference talks, and Q&A sessions grounded in real production experience: agentic systems, RAG pipelines, eval design, observability, and team upskilling. If you are planning a technical event and want someone who can field the hard questions, reach out to discuss your event's goals and audience.