Heading to Ai4 After Stanford and Berkeley: The Questions I Want Answered
4 min read
Lance Ennen
Share:
I'm heading to Las Vegas for Ai4 2026 — August 4–6 at The Venetian, billed by its organizers as America's largest AI industry event: 12,000+ attendees, 1,000+ speakers, 400+ exhibitors, and a headline keynote that puts Geoffrey Hinton, Fei-Fei Li, and Andrew Ng on one stage Tuesday morning.
It's the final stop of a week that already reshaped my thinking twice. At Stanford BASS, I watched Visa lay out a stablecoin strategy that has quietly become a $7B-annualized settlement operation. At the Berkeley Agentic AI Summit, I watched researchers and builders converge on the same lessons my team learned running five development agents in production: harnesses matter more than models, and security is the quiet headline.
Stanford was the money. Berkeley was the machinery. Ai4 is the market — the place where enterprise buyers, not researchers, decide what actually gets deployed. Here's what I'm going to find out, stated before I arrive so you can hold me to it.
The image above is an original illustration, not an event photo. Real photos start when the doors open — follow the live coverage hub.
The questions I'm carrying
On agent orchestration. Startups orchestrate agents with open tooling and glue; my own stack is OpenClaw plus discipline. What are enterprises actually using? Is there a real enterprise control plane emerging, or is it consultants and slideware?
On production deployments. Every vendor claims Fortune 500 agents in production. I want specifics on stage: which workflows, what autonomy level, what intervention rate, who reviews what. "Deployed" spans everything from a chatbot in a sandbox to agents with commit access — the distance between those is the whole story.
On memory. Long-running agents live or die on memory and context management. Berkeley treated it as an open research problem. Do enterprise deployments have an answer, or are they scoping around it with short tasks?
On observability and evaluation. My team's biggest gap, and I suspect everyone's. Who has evaluation running like CI? What does an agent incident postmortem look like at a bank?
On permissions and approval workflows. The control-plane thesis says approval gates and policy engines are where autonomy becomes safe. Enterprises are the natural buyers. Is anyone selling — and is anyone deploying — real policy enforcement, or is "human in the loop" still a checkbox?
On AI software engineering. Agent-written code is normal at startups. In the enterprise, who's merging agent PRs? Under what review regime? What happened to their defect rates?
On digital identity. Agents acting on behalf of companies need identity, attribution, and delegation chains. This is the least-discussed prerequisite for everything else. I'm looking for anyone treating agent identity as seriously as employee identity.
On payments and stablecoins. The convergence I saw from the Stanford side — agents that move money over programmable rails — needs the enterprise side to show up. Will payments infrastructure even have a presence at Ai4, or are these still two separate conversations?
Five predictions, on the record
I'm putting these on the prediction scorecard now, before the conference. They get scored during the event — confirmed or busted, never rewritten:
- Agent orchestration will be the loudest theme on the expo floor — more booths selling coordination, observability, or governance than model access itself.
- Most enterprise "agent" deployments presented on stage will turn out to be single-workflow automations with a human approval gate, not multi-agent systems.
- Memory and context management will be named the top unsolved problem by at least three speakers across different tracks.
- Stablecoins and AI-driven payments will barely register at Ai4 despite dominating Stanford — the two worlds haven't merged yet.
- At least one major vendor will announce an "AI employee" product whose demo quietly depends on human review for anything consequential.
If I go five-for-five, the week's thesis holds: the technology is ahead of the governance, and the governance is ahead of the enterprise's willingness to admit what "deployed" means. If I'm wrong — especially about #4 — that's genuinely better news than being right.
How to follow along
Everything lands on the Ai4 2026 live coverage hub: daily recaps, keynote notes, startup discoveries, product launches, a company of the day, and the scorecard above. After the week ends, the series closes with a synthesis — three conferences, one conclusion.
See you in Vegas.
I'm attending Ai4 as an independent publisher; my attendance is self-funded and I have no commercial relationship with the event. Predictions above are opinion, clearly labeled. See the editorial policy.
Written by
Lance Ennen
CTO & Technical Advisor helping startups and Fortune 100 companies build innovative digital products. Passionate about blockchain, AI, and scalable architecture.
Enjoyed this article? Share it with others