Harness Era AI-First Organizational Structure

This episode of Silicon Valley 101 invited the three founders of Creao, focusing on Harness Engineering. It’s not simple Prompt Engineering, nor Context Engineering — it’s about truly treating AI as the primary productivity entity, redesigning the entire company organization and workflow.

From Prompt Engineering to Harness Engineering

In the past three years, large model engineering capabilities have gone through three evolutions.

In 2023, people were talking about Prompt Engineering, focusing on how to write good prompts; in 2024, it became Context Engineering, focusing on how to provide more complete context to the model; by 2026, a new term suddenly emerged: Harness Engineering.

Harness originally means the straps put on a horse, now used to describe: how to build a system around large models that can self-heal, self-improve, and continuously work in the real world.

Real AI-First Is Not Treating AI as an Auxiliary Tool

Creao’s CTO Peter previously worked in Meta’s Llama team and Apple’s multimodal group. He tweeted a post that directly exploded with 1.87 million views.

The core point is: most companies’ so-called AI-First is fake. Real AI-First transforms AI from an auxiliary tool into the dominant productivity driver, completely restructuring organizational architecture and workflows.

The result: they write a feature at 10 AM, run an AB test by noon, kill it by 3 PM based on data, and rewrite a better version by 5 PM.

The traditional process might take six weeks; now it’s done in one day.

The Most Shocking Part Is That Roles Are Completely Reversed

Previously, product development was slow, so marketing teams could prepare four to five months in advance. Now it’s the opposite: development speed far exceeds marketing, and marketing teams are chasing features that engineers ship.

Clark says they no longer need feature wishlists or bug lists. Because bugs can be detected and automatically fixed by agents, and there are way too many features to ever use.

This change is huge.

In the past, the scarcest resource was “development capacity”; now it’s “whether the organization can digest AI capacity.”

Trust Shifted from “Trusting People” to “Trusting AI”

Another major change: the way cross-team collaboration works has changed.

Before, cross-team alignment had extremely high costs, with engineering, product, and marketing needing constant synchronization. Now AI can directly tell Marketing: what Engineering will ship today, without human back-and-forth.

Peter and his team even dismantled the product manager role.

Because in their view, alignment cost itself is a false need. As long as the system is strong enough, decisions can be AI-driven, and information can flow automatically through AI.

Junior Engineers Find It Easier to Adapt to AI-First Environments

Even more counterintuitive: junior engineers adapt to AI-First environments more easily than senior engineers.

Senior engineers have too much technical debt and mindset inertia, unwilling to expand their scope to product design and data analysis. Junior engineers are more willing to cross functional boundaries.

Peter says the most valuable people in the future may not be deep specialists in a vertical field, but generalists who simultaneously possess Architecture + Product Sense + Marketing Sense.

This is worth pondering.

Once AI amplifies execution capabilities, the truly scarce resource becomes: can you define problems, judge direction, understand users, read data, and set up the system?

Agent Economy: Future Content May Not Be for Humans

The part that made my head spin is Clark’s mention of the Agent Economy.

They’ve found that many marketing materials today are actually for agents, not for humans.

In the future, buying things, subscribing to newspapers, or ordering milk may all be decided by agents. So the content you produce today — is it for human consumption or for AI consumption?

The value evaluation criteria might completely change.

In the past, we optimized for human attention; in the future, we may also need to optimize for agent understanding, retrieval, judgment, and action.

Harness Is Divided into Two Layers

They split Harness into two layers.

The first layer is the Harness for the company’s internal agent system, focusing on self-healing and auto-fixing.

The second layer is the agents built by users on the Creo platform, which can also continuously self-improve.

Their entire team is only 25 people, with fewer than 10 engineers, yet they completed the entire architecture reconstruction in two weeks.

Peter says a year ago, a product of this scale would have required at least 100 people working for four to five months.

Of Course, the Cost Is Real

This is not to say the AI-First transformation has no cost.

They spent a long time aligning the mindset of the entire team, especially making everyone truly believe that AI can lead, not just assist.

Many people say they embrace AI, but essentially they only want to use AI tools to improve their own efficiency, unwilling to cede the productivity driver role to AI.

This mindset gap determines whether a company can truly complete an AI-First transformation.

Biggest Cognitive Upgrade

After listening to this episode, my biggest cognitive upgrade is:

AI-First is not about layering AI tools on top of existing processes, but treating AI as a new species and redesigning the entire organizational ecology.

The core value of humans in the future is not execution or alignment, but defining the direction of requirements and ultimately reviewing whether the results align with human interests.

This episode has extremely high information density. The full transcript is very long — I recommend going to Podwise for the complete version.

What do you think — will the product manager role disappear, or will it evolve into a new form?

Feel free to discuss in the comments.

More Hardcore Podcast Content: Podwise @PodwiseHQ