---
title: "Welcome Back - Featured Keynotes"
speakers: "Alex Lieberman (Tenex)"
duration: "4:54"
source: transcript.md
video: "https://www.youtube.com/watch?v=akEqiXFo3GM&t=24163s"
themes: [bounded-autonomy-runtime-governance-and-the-enterprise-diffusion-gap]
kind: talk-story
status: draft
---

# Ninety percent of enterprises rolled out an LLM. Five percent run agents outside engineering.

Opening the afternoon keynote block on Day 1, Tenex co-founder Alex Lieberman shared field numbers from roughly a hundred large enterprises across financial services, healthcare, and industrials. Access to chat models is nearly universal, yet production agents outside software engineering have barely started.

## Opening the final Day 1 keynote block

Lieberman brought the Simons Room back for the closing keynote stretch of Tuesday afternoon, pausing first to thank presenting sponsor Arize alongside platinum sponsors Oracle, Sonar, WorkOS, and Exa, plus the supporting partners behind the conference (00:00:00). Rather than treating the host slot as pure logistics, he used the five-minute reset to report what he had been hearing offstage between sessions (00:00:00).

**Key points**

- The segment opens the final afternoon keynote block on Day 1 in the Simons Room.
- Lieberman credits Arize, Oracle, Sonar, WorkOS, Exa, and supporting event partners before turning to field notes from enterprise deployments.

> “Well, welcome back for the closing keynote block here at AI Engineer New York 2026.”
> — 00:00:00

**Visual:** slide `experience/stage-opening.jpg` (main stage wide view opening the afternoon keynote block)

## Stage talks are shared surface area for hallway conversations

Standing in the back of the room during earlier tracks, Lieberman noted that the real function of dense technical talks is to create shared context for the people listening (00:01:35). Even when speakers share deep implementation details from the frontier, the presentation itself is mainly a starting point for practitioners to compare notes in the hallways afterward (00:01:35).

**Key points**

- Technical talks work best as shared surface area that starts peer-to-peer conversations between sessions.
- Offstage hallway threads centered on model trade-offs, hiring AI engineers at scale, and the gap between frontier demos and enterprise rollout.

> “The depth by which they talk about the work they're doing is incredible, but this is also just surface area that acts as a starting point for conversation amongst everyone in the room”
> — 00:01:35

**Visual:** slide `experience/tenex-diffusion-talk.jpg` (Alex Lieberman on the Simons Room stage)

## Four milestones of enterprise AI diffusion: 90%, 60%, 50%, 5%

Recounting a backstage conversation with conference co-organizer swyx (transcribed as "Switz"), Lieberman laid out how AI adoption drops across roughly a hundred large enterprises Tenex advises in financial services, healthcare, and industrials (00:02:29). He framed enterprise rollout as a four-milestone race. First, roughly 90% of those organizations have rolled out a general-purpose LLM to employees (00:02:29). Second, about 60% have trained staff on how to apply those interfaces in daily work (00:02:29).

The steeper drop arrives with autonomous workflows. Around 50% have put coding agents into active engineering and product delivery, making software teams the clear beachhead (00:02:29). Outside engineering, production agents working in business workloads sit at roughly 5% (00:02:29). Alongside open-source versus fine-tuned model choices and the difficulty of hiring AI engineers at scale, closing that 90-to-5 gap defines the work of the next 12 to 36 months before Lieberman hands the stage to Coatue's Frank Long (00:02:29).

**Key points**

- Across ~100 large enterprises Tenex advises, ~90% have deployed a general-purpose LLM and ~60% have trained employees to use it.
- Roughly 50% run coding agents inside engineering and product workflows.
- Only ~5% run production agents in non-engineering workloads, leaving a 12-to-36-month enterprise diffusion gap.

> “Next up is coding agents, agents that are actually used within engineering and product workflows — that's around 50% of all the enterprises we've been in, and then the final one where we are truly still day one is agents actually working in non-engineering workloads in an organization, that is around five percent right now.”
> — 00:02:29

**Visual:** diagram "Four-stage enterprise AI diffusion drop-off across ~100 large firms: 90% LLM access to 5% non-engineering agents"