AI Is Moving From Answers to Outcomes

Anthropic’s latest Economic Index report contains a finding that caught our attention: 93% of Claude conversations now produce an identifiable artifact.

That might sound like a small change in how people use an AI assistant, but we think it points to something much bigger. People aren't just using AI to ask questions and get answers anymore. Increasingly, they're using it to produce something they can actually take away and use: a document, report, presentation, analysis, plan, piece of code or another concrete output.

For work-related conversations, documents and reports are the most common type of output in Anthropic’s data, accounting for 20% of conversations. At the same time, Claude is increasingly being used for more complex and autonomous work, particularly through products such as Claude Code and Cowork. Anthropic's data shows that these more autonomous tasks also tend to involve more compute, reflecting the additional reasoning and work required.

The direction seems fairly clear. AI is moving beyond simply answering questions and towards participating in the work itself.

The model is getting better. So what happens to the software around it?

This creates an interesting challenge for the companies building AI software today.

A lot of early AI products were built around the idea that the model itself was the differentiator. They took capabilities from a foundation model, wrapped them in a specialised interface and gave customers a reason to use their product instead of going directly to ChatGPT or Claude.

That made sense when the models were less capable and the interfaces around them added significant value. But the foundation models are improving rapidly. Claude can already analyse transcripts, work across documents, reason about business information, create reports and produce increasingly sophisticated outputs.

So a reasonable question for any AI application is becoming: what happens if Claude gets much better at doing the thing your product currently does?

We think that's an important question, and not necessarily a bad one.

At Productised, we're deliberately taking a different approach. We don't want to compete with Claude's intelligence or pretend we've invented something the underlying model can't do. We're not just putting a nicer interface around something Claude already does.

Instead, we see an opportunity to build the layer around that intelligence that makes it useful to a particular business and its customers.

Claude provides the intelligence. Productised adds the business context, expertise, experiences, lead capture, branding and personalised outputs that turn that intelligence into something a business can actually put in front of a prospect or client.

That distinction becomes more important as the models improve.

Expertise still matters

Another part of Anthropic's report is particularly relevant here. When users were asked what AI still struggles with, experienced workers were more likely to identify things such as judgment, contextual awareness, situational reasoning, trust and interpersonal relationships. People with more than 15 years of experience were especially likely to say that the accumulated, context-specific expertise they have developed is difficult for AI to replicate.

We don't see that as a reason to keep AI away from professional expertise. Quite the opposite.

A consultant has a methodology. A coach has a framework. An agency has a way of diagnosing a client's problems. A professional services firm has years of knowledge about what good looks like and how to decide what a client should do next.

The interesting opportunity isn't simply putting all of that knowledge into an AI and asking it to reproduce the expert.

It's giving the AI enough context to apply that expertise to a particular situation.

A prospect might complete an assessment, provide information about their business, answer a series of questions or share details about a particular problem. The business brings its own methodology, criteria and knowledge into the experience. AI can then reason across the two and produce something that is relevant to that individual.

That's a very different proposition from simply asking Claude to write a report.

From information to something useful

Consider a consultant who has developed a framework for assessing business readiness.

Claude can already help that consultant turn their framework into questions, analyse responses and write a report. They don't necessarily need another AI model to do that.

But there is a bigger product opportunity around the process.

A prospect could arrive on the consultant's website, work through an interactive experience, receive a personalised diagnosis based on their answers and get recommendations for what they should do next. The consultant receives the prospect's information and a much better understanding of their situation before they ever have a conversation.

The same underlying capability could be used by a coach to create a personalised action plan, by an agency to build a client diagnostic, or by a professional services business to create a calculator or recommendation tool.

The important part isn't the format of the output. It is the connection between what the business knows, what the individual tells it, and what AI can work out from the two.

That is the problem we're interested in solving at Productised.

The interface for AI is changing

Anthropic's research also gives us a useful clue about where AI software is heading. The way people interact with a model increasingly determines what they are able to delegate to it. Claude Code, for example, enables a much higher degree of autonomy than a traditional chat interaction, even when the underlying model is the same.

That suggests the future of AI won't simply be one enormous chatbot where everyone manually prompts a model to do everything.

There will be different layers around increasingly capable models. The model provides the underlying intelligence, while applications provide the context, workflow, constraints, integrations and experience that make that intelligence useful for a particular job.

The business then brings its own expertise and intent into that system, and the customer receives the resulting outcome.

That's the role we're building towards with Productised.

Why we're building around Claude

This is also why we're comfortable being very transparent about our relationship with Claude.

We don't think businesses should have to choose between using Claude and using Productised. If Claude is where a business already keeps its knowledge, works through problems and increasingly gets work done, we'd rather extend that environment than create another AI silo.

Our Anthropic-verified connector brings Productised into Claude for lead generation and client delivery. Claude provides the intelligence; Productised provides the business-facing layer that turns that intelligence into branded, personalised experiences for prospects and clients.

And the relationship works in both directions. As Claude becomes more capable, the intelligence available inside Productised becomes more capable too. Customers don't need to replace their platform every time the underlying model improves. They can continue building on the same system while benefiting from improvements in the model underneath it.

We think that's a much healthier relationship with foundation models than trying to convince customers that they should stop using them.

The opportunity isn't another AI

This is probably the biggest takeaway we have from Anthropic's latest report.

As foundation models become more capable, building another AI that can answer questions, write reports or analyse information becomes less interesting on its own. Those capabilities are increasingly becoming infrastructure.

The more interesting question is what businesses can build because those capabilities now exist.

For Productised, that means helping businesses turn their expertise and their customers' information into useful, personalised experiences. It means giving prospects something valuable before asking for their time. It means helping professionals apply their knowledge at scale without removing the human expertise that makes their service valuable in the first place.

We don't need Claude to get worse for Productised to become more valuable.

We want the opposite.

The better Claude becomes at understanding context, reasoning and producing useful work, the more capable the experiences businesses can build with Productised become.

That's the opportunity we see in the next generation of AI software: not competing with the models, but building the business capabilities that make increasingly powerful models useful in the real world.

Claude thinks. Productised delivers.

Keep reading