Article · 2026-07-14 · 6 moments

5 Trends That Defined AI Engineering at World’s Fair 2026

At this year's AIE World’s Fair, AI engineering entered a new phase: building systems around agents, rather than just building with agents. ✦ AI generated

Read original ↗

GUEST Thariq Shihipar GUEST Lilian Weng GUEST Peter Steinberger GUEST Dex Horthy GUEST Zach Lloyd GUEST Paul Bakaus HOST Shawn Wang (swyx)HOST Alessio Fanelli

Claim◆

Current enthusiasm and hype around fully autonomous agent loops has outrun the actual engineering discipline needed to make such loops reliable, unlike Kubernetes-style loops which are deterministic.

In the AIEWF loop debate, Dex Horthy argued that hype around autonomous agent loops has outpaced real engineering discipline, contrasting them with deterministic control loops like Kubernetes.

transcript

Dex Horthy: Dex Horthy from HumanLayer claimed that “the hype is outrunning the discipline.” He wasn’t against loops, per se, noting that Kubernetes is built on control loops — “but they’re deterministic loops.”

Mechanism

Frontier models like Claude are more like organic systems that are grown rather than designed, exhibiting a 'capability overhead' where intelligence increases unpredictably and spikily rather than smoothly.

Anthropic's Thariq Shihipar described Claude Fable as an organic system whose capabilities emerge unpredictably, growing 'in a spiky way' rather than through controlled design.

transcript

Thariq Shihipar: Anthropic’s Thariq Shihipar talked about how their latest model, Claude Fable, is like an organic system — “models are grown, not designed.” There’s a “capability overhead,” he said, where “Claude gets smarter in a spiky way.”

rebuts · 1supports · 1

Claim

Most AI skills, like most models, are not very creative — they converge in a single direction, so when everyone uses the same skill for tasks like frontend design, the outputs end up looking homogenized.

Impeccable creator Paul Bakaus warns that AI skills tend to converge outputs in one direction, so shared skills risk making designs and outputs look increasingly the same across users.

transcript

Paul Bakaus: most skills — and indeed most models — are not very creative. “They converge in one direction, and if everybody uses the same skill to do frontend design work or something like that, everything ends up looking the same,” he said.

Example

Organizations building 'software factories' must deliberately choose which parts of the software lifecycle to automate and which points require human review, since preferences differ across organizations and codebases.

Warp CEO Zach Lloyd explains that enterprises adopting agentic 'software factory' platforms like Oz must decide case-by-case which repositories and lifecycle stages to automate versus keep humans in the loop for.

transcript

Zach Lloyd: You choose your repositories, the parts of the software lifecycle you want to automate, and the points where humans should be brought into the loop. Different organizations and codebases will have different preferences. Do you fully automate code review? Do you have humans review certain high-risk changes?

extends · 1provides context · 1

Claim

The human engineer's proper role is to set direction and make decisions in an 'outer loop' while letting the agent autonomously run the 'inner execution loop.'

OpenClaw creator Peter Steinberger describes deliberately positioning himself in the 'outer loop,' directing and deciding, while the agent executes the 'inner loop' of actual work.

transcript

Peter Steinberger: Even the “ClawFather” Peter Steinberger, creator of OpenClaw, makes a point of putting himself in the outer loop. In the OpenAI keynote, he explained that “the agent runs the inner execution loop; I set the direction and I make decisions in the outer loop.”

gives example · 1

Claim◆

The system surrounding an AI model — the harness that manages workflows, context, permissions, evaluation, persistent state and continuous improvement — has become just as important as the agent itself, marking a shift from prompting models to engineering reliable systems around them.

Lilian Weng's 2026 essay argues that AI engineering has moved past building autonomous agents toward engineering the harnesses (workflow, context, evaluation, state) that surround them.

transcript

Lilian Weng: Rather than focusing on the agent itself, Weng argues that the system surrounding the model has become just as important: the harness that manages workflows, context, permissions, evaluation, persistent state and continuous improvement. In other words, AI engineering has moved beyond prompting models toward engineering reliable systems around them.

extends · 1gives example · 1supports · 1

Highlight slides

Hype Has Outrun Engineering Discipline✦ from: Current enthusiasm and hype around fully autonomous agent loops has outrun the actual engineering discipline needed to make such loops reliable, unlike Kubernetes-style loops which are deterministic.The Harness Is the New Frontier✦ from: The system surrounding an AI model — the harness that manages workflows, context, permissions, evaluation, persistent state and continuous improvement — has become just as important as the agent itself, marking a shift from prompting models to engineering reliable systems around them.Deterministic vs. Autonomous Loops✦ from: Current enthusiasm and hype around fully autonomous agent loops has outrun the actual engineering discipline needed to make such loops reliable, unlike Kubernetes-style loops which are deterministic.From Prompting to Systems Engineering✦ from: The system surrounding an AI model — the harness that manages workflows, context, permissions, evaluation, persistent state and continuous improvement — has become just as important as the agent itself, marking a shift from prompting models to engineering reliable systems around them.

Related episodes

Video · The Cognitive Revolution

Current enthusiasm and hype around fully autonomous agent loops has outrun the actual engineering discipline needed to make such loops reliable, unlike Kubernetes-style loops which are deterministic.

Frontier models like Claude are more like organic systems that are grown rather than designed, exhibiting a 'capability overhead' where intelligence increases unpredictably and spikily rather than smoothly.

Most AI skills, like most models, are not very creative — they converge in a single direction, so when everyone uses the same skill for tasks like frontend design, the outputs end up looking homogenized.

Organizations building 'software factories' must deliberately choose which parts of the software lifecycle to automate and which points require human review, since preferences differ across organizations and codebases.

The human engineer's proper role is to set direction and make decisions in an 'outer loop' while letting the agent autonomously run the 'inner execution loop.'

The system surrounding an AI model — the harness that manages workflows, context, permissions, evaluation, persistent state and continuous improvement — has become just as important as the agent itself, marking a shift from prompting models to engineering reliable systems around them.

Fable's Back, AI Engineer Recap, & SambaNova

Software in the Age of Agents | The a16z Show

Is Defense the Next Trillion-Dollar Category? | a16z American Dynamism Summit