It’s fair to say that AI has now mostly solved code generation. Leading models max out every software engineering benchmark. And agent-native companies are starting to see a real impact on engineering throughput, up to 10x from Anthropic to Theory portfolio company Artemis. 

The most common prediction is a SaaSpocalypse: if software is nearly free to build, the value of software products collapses. Competitors (or customers) can clone any feature. Agents will generate a design, write the code, and maintain it, all with barely a human in the loop. Differentiation goes to zero.

We think the conclusion is mostly backwards. In a world where software engineering gets faster and cheaper, product judgment will matter more than ever. The scarce skill isn't writing code. It's deciding what to build and aligning the company around delivering it.

Good Code Isn't Good Product

With coding agents, we can produce a near-infinite amount of quality code. But a near-infinite amount of quality code does not make a near-infinite amount of quality product.

First, every product is built on a series of core strategic assumptions. Do you build headless or UI-first? Design around synchronous or asynchronous workflows? Each has trade-offs, and every decision you make moves the platform down a path that gets harder and more expensive to reverse. Agents can write and rewrite code, but still accumulate the weight of thousands of commits that could have been built in the wrong direction.

Second, there are non-engineering limiters to new feature velocity. AI-supercharged engineering teams might ship dozens of new features every day. How do you teach your customers how to use them? What about your sales and support teams? How do you know what’s working or not?

With unlimited engineering capacity, product bottlenecks don't disappear. And as companies (and their products and user bases) grow, the structural and organizational inertia does too. That makes the product decisions of what, when, and how to ship even more critical – when everyone can build anything, what you choose to build is the whole game.

Can’t AI Make the Product Decisions?

Why not just give Claude your customer feedback and have it decide what to do?

For simple feedback and bug fixes, sure. But AI is inept at the two key things that define good product judgment.

• Models don’t know how to say “no”. The best product companies are defined as much by what they didn’t build as what they did. Any PM knows that if you build everything every customer asks for, you’ll end up with a bloated, disjointed product that serves nobody well. But give an LLM a laundry list of customer requests, and it will produce a laundry list of features to build.
• AI is good at the norm, not the frontier. It can optimize a checkout page because it has seen a million of them. But ask it to design a novel kind of real-time assistant and it will struggle. The most valuable product decisions push the frontier of what exists today, and that’s where models are weakest.

Will better models close these gaps? Certainly, but both weaknesses are largely structural. LLMs are trained to complete user requests and don’t have the right context to determine which to listen to or not. By definition, they learn from the distribution of software that already exists.

And critically, most important product judgment is not verifiable – they are strategic decisions that might take months or years to play out. Until LLMs can lead a whole company fully autonomously, we think human judgment will remain necessary.

What New Capabilities Will Product and Engineering Teams Need?

Today's product and design processes were built for a world where engineers were the scarce resource, and everyone else filed requests into their queue. That is changing. Product leaders should rethink not just their tooling but also how decisions are made for this new paradigm.

Here are four themes we’re interested in. We’ll share more detailed thoughts about each.

1. Collaboration platforms to manage agent-velocity work

Agents will be able to produce far more work than humans have time to review. Using today’s tools like Slack and Linear, people will find themselves drowning in content to review, and agents will struggle to find the right context. We will need new kinds of platforms that allow many humans + agents to align on a strategic direction, direct attention to the most important decisions, and collaborate directly on rich artifacts or prototypes instead of walls of text.

2. Agentic experimentation & simulation platforms

Building hundreds of new features in parallel will break today’s experimentation and analytics platforms, which are designed for occasional A/B tests and ad hoc data analyses. The next generation of these tools will need to be AI-native, allowing agents to autonomously design, run, and interpret experiments, then update software accordingly. They will need to be designed around both human and agent users, as agents come to represent a majority of software usage. And they will increasingly rely on simulation to test variants pre-launch.

3. Automated product research & discovery

Even when a product decision is best made by humans, there is a lot of work going into them that can be automated by agents. The next generation of PMs will have agents continuously mining customer requests and sales calls for insights, running user interviews, and synthesizing evidence to inform the product roadmap.

4. Change management & education

The pace of new feature launches will overwhelm users. We think that a majority of education will need to shift to the product itself. An in-app assistant can directly tell the user what they can do, or the UI might highlight a new feature that it thinks a customer would use.

The same issue will make sales and support reps’ lives challenging; they may need contextual information on new capabilities surfaced real-time in sales calls or support requests.

—

The SaaSpocalypse gets one thing right. Cheap code does collapse the value of writing code. But it raises the value of everything code can't do for you: knowing which product to build, which tradeoffs to accept, and rallying an organization to deliver it to customers. For product and engineering leaders, the opportunity is not simply to ship more. It is to build a stronger decision system that turns faster execution into better customer adoption.