Coté

AI companies are building platforms for running agentic applications. Right now, those applications are primarily for software development, with a little bit of knowledge worker stuff. In each case, you get a “harness," an application that wraps all sorts of functionality around a model.

This harness app is way beyond the chat-based apps we grew up with over the past few years. They use the model to figure out multi-step processes and get access to data and other apps - accessing files, working with your email, PowerPoint, etc.

What using “harness” looks like is clear with programming: you’re using AI in each step of the software development lifecycle. You want to create a new application, or add features to an existing one. The harness comes up with what the features and changes should be, planning it out. It structures the dependencies, writes the code, compiles and packages it, runs it to test it, debugs the app, and then deploys the app to wherever it runs.

What this means for other types of work is less clear and hasn’t really been solved well. There’s plenty of AI features built into existing apps: little sidebars in Google Docs, pretty good sidebars in video editing, and of course replacing customer service chat sidebars.

All that “agentic AI” hype from late 2024 through 2025 was real - remember when McConaughey showed up for dinner soaked in rain? - but it took until late 2025 for it to start becoming real and get real usage. I’d think of the “agentic AI” discussions we’re having now as harness building and use-exploration.

I don’t think we’ve figured out what a harness looks like for non-programming work. You see plenty of engineering types who’ve figured out how to get a list of their meetings today, track tasks, and other “admin” things. These are new skins on productivity apps, which tend to be great. They’re giving people more agency. It’s all bespoke, one-off stuff at the moment, though. That’s not cool for large organizations. What are the harnesses that cover the full lifecycle of various knowledge worker tasks?

The AI companies have been buying up industry-focused companies that, we have to guess, are figuring out the equivalent of Claude Code for the medical industry, for accounting. What does a harness for paying your taxes look like? For ongoing financial management and planning? It can’t just be a sidebar in the app that has access to your data, a handful of skills and MCP servers. It’s got to be something different.

At Tanzu we’re in the platform business, not the app business. We announced the new version of our platform yesterday, Tanzu Platform 10.4. It puts several of the building blocks in place to run the backend of these harnesses: identity and audit trails for agents, governed access to models and MCP servers, a gateway that controls what tools agents can call, deny-by-default networking so agents can’t wander into things they shouldn’t, and cost tracking so you know which agent is burning through your token budget. To me, the main things this kind of enterprise platform delivers are security, compliance, and consistency. You’re also looking for ways of standardizing how teams of people - thousands of people, eventually - use your software. You know, all that enterprise-y stuff.

This matters because large organizations are trying to figure out how to use AI and building out these apps. Or, they should be. They should be figuring out how to build harnesses, integrations, and all the wiring that goes into customizing AI into how they currently work. They need a backend to run parts of those AI-driven workflows, the backend for the harnesses they’ll need.

And, you know, also just run their applications.

So, what we have in Tanzu Platform 10.4 is our point of view of what that looks like in an enterprise stack. We’ve built platforms that do this for traditional applications for 15 years now, starting with Cloud Foundry in 2011. The Tanzu Platform (which uses Cloud Foundry as its base) is an application platform, a PaaS. And what we’re doing here is extending that platform for AI applications. There’s a private cloud angle here, of course: you can use public AI services and public cloud PaaSes, but with ours you can run it on your own infrastructure.

Data is always a mess, so you’re always tidying it up

There’s the data angle here too. It’s a cliche now that enterprise AI use needs data to be valuable. Sure, but this just leads to the next question: how do we get access to it. The secret of enterprise data is that it’s always a mess and probably will always be. The floor for data architecture is always rising - once you get it cleaned up and perfect, you finally get the chance to start doing new types of analysis, more frequently. You allow more people to use it, they come up with more scenarios, they need to add in more data, new apps come along that have their own data format…and you start cutting corners in your glee for finally having usable data and start using million dollar rolls of duct-tape to slap together data stacks.

This is re-discovered by every new entrant in the data and analytics space. Here’s a candid take on that from Palantir folks:

I think the critical part of it was really realizing that we had built the original product presupposing that our customers had data integrated, that we could focus on the analytics that came subsequent to having your data integrated. I feel like that founding trauma was realizing that actually everyone claims that their data is integrated, but it is a complete mess and that actually the much more interesting and valuable part of our business was developing technologies that allowed us to productize data integration, instead of having it be like a five-year never ending consulting project, so that we could do the thing we actually started our business to do. Shyam Sankar, then CTO at Palantir, 2023.

Data is very important to figuring out these harnesses. It’s too easy to dismiss it, or focus on just parachuting in forward deployed engineers to build up a datalake of meat-sacks.

Anyhow.

A private cloud platform for harnesses (and, as always, plain old apps)

That’s how I’d look at what Tanzu Platform 10.4 is: our vision of a private cloud AI stack. Pretty much every large organization wants and needs this - you should check it out!

Here’s a round-up of announcement blogs, posts, etc. to get more into it: