Introducing, Komo AI (aka Kai)
Thoughts as we launch our first big suite of AI capabilities
I’ve been talking, experimenting, sharing and thinking a lot about AI recently. In fact, it’s why I started this Substack. I’ve said it before, but I don’t think there has ever been a more exciting time in technology and startups, at least not in my 15 years in the industry.
Until now, I haven’t spoken about what I’ve been building, or we’ve been building more accurately. One reason is, once you really go deep on the tools, you do realise that so much of AI is just marketing. The skeptics are right in this regard. A lot of the tools are a total letdown. The demo is often better than the product.
There are good reasons for this, building reliable and high quality AI-powered products with probabilistic LLMs is not that easy, it turns out. Experimenting is very easy, getting it right, is much harder.
There are plenty of exceptions, to be clear. Posthog, Notion and Linear are just three examples of world-class product organisations using AI to make their products better, move faster, and change the way they work internally.
I have been extremely conscious about our first AI product being a let down, so, we’ve intentionally kept it quiet. I know this goes against so much of the traditional “if you’re not ashamed of your V1, you’ve launched too late” ethos in startups, but somehow, AI feels different. If it’s not good, it’s so easy to dismiss it as hype, as jumping on the sparkle bandwagon, and missing the point.
We’ve been building with some pilot customers for the last few months, trying to get a really good V1 ready. And today we’re ready to share it with the world. It’s by no means perfect, we’re still very much in the early days, and like every other product team, trying to think through the big AI product topics: evals, observability, quality, context and memory.
Our thesis around Komo AI is very simple. We think there is tremendous value in collapsing the time required to ideate, build and launch content and campaigns within our platform, from hours to minutes. The way we’ve thought about this is, by automating a lot of the “scaffolding” work in the software, literally the points and clicks, we can help our customers get the work done, a whole lot faster.
We wanted to go beyond, and do more than a chat interface, and make Komo AI feel like a real coworker. Another marketing coordinator, in your browser, ready to work for you.
We are launching with two core product capabilities.
Firstly, “Campaign Ideation”. This helps marketers generate campaign ideas, respond to briefs and just generally ideate about what they could be building. Is this just a ChatGPT-wrapper? I mean, in a way yes, but the benefit is, every idea is able to be executed within the Komo product. We have built a strict set of schemas to ensure that everyone Komo AI throws back at you, you could pull off. In this sense, it’s a far more helpful way of ideating than just using ChatGPT or Gemini.
Secondly, Komo AI can handle “Campaign Creation”. For any brief, Komo AI will create and scaffold out the content, in our case, the Card for you. The real value here is just how much time is saved. Komo is a fairly deep product, used by enterprise consumer brands to build promotions, competitions and campaigns that reach hundreds of thousands of their customers. Collapsing the create time (the actual working-in-the-UI-creating-and-clicking work) for a campaign from hours or even days to minutes, is actually pretty magical.
We’ve gotten great feedback and validation so far from marketers at some of the world’s biggest consumer brands.
Here’s a full demo of what we’re launching and these two capabilities:
So we’re proud of the V1, but excited to keep building and iterating. Here are some of the bigger things on my mind as we look to improve the product with customers:
Broader “capability” across the entire product footprint. We want Kai to be able to reach into the deepest corners of the Komo product and do work. Changing settings, looking at integrations, detecting misconfiguration or errors etc.
Connect our ideation and creation tools. Currently these are disconnected experiences. We’ll look to bring these together so you can float between ideating and creating in the same workflow. I’m very inspired by how the “Canvas” tool works in ChatGPT, and the coding agents like Cursor and Claude Code provide interesting new ways to think about building and ideating" at the same time.
Provide more AI tools “inline”. Currently, our AI tools are all about the creation step. Scaffolding out a draft. Getting you to 80% complete. Where we’ll look to next, is providing more tools once content has been created, for editing and refining. I guess you could call this like the ambient co-pilot for want of a better term.
Building more AI tools for insights and analytics. One of the core primitives in our product, like most enterprise SaaS products is analytics and reporting. I’ve been particularly impressed with Posthog’s “Max” here, and how easy Max makes it to generate complex SQL queries, interrogate the data and play around with different charts and visualisations. Perfect example of collapsing the complexity curve inside software.
Exploring MCP. MCP is very early, still far too unintuitive still for most users, but is by far one of the most interesting emerging concepts in AI for enterprise. I could see us exposing an Komo MCP server that lets customers extract data, insights, build insights and generally ask questions of their Komo campaigns wherever they spend most of their day, increasingly I think that is ChatGPT, Copilot, Gemini or Claude for many marketers.
Asset/Image generation. We’ve really not gone deep here yet, given for our customers (big consumer brands), AI-generated imagery is just not there yet. But things are happening, and quick. So I’m intrigued to see what could be done in the “remix” space, given the latest developments from Google here with their nano banana image editing. If a customer gave us enough reference material, could we create net-new assets that stay true to exiting assets?
Deep Research. It’s become a fairly standard capability at this point, but for good reason. Being able to use the latest reasoning models, and run it over every campaign, every dashboard, every interaction inside a customer’s instance, and using that to build reports, fix issues or suggest improvements seems really interesting and exciting.
It’s an exciting (and kind of insane) time for building B2B software. All the rules are being re-written in real time. So we’re excited to be leaning in. But for all that said, as much as things change, so much stays the same. Keeping the customer front-of-mind, building intuitive products that are joyful to use and optimising for the customer’s desired business outcomes and results. Those principles haven’t changed, and I suspect they never will.
Read more about Komo AI on our blog post: https://blog.komo.tech/meet-kai-create-campaigns-in-minutes-while-competitors-wait-weeks