In partnership with

If you’re a keen follower of our quarterly roundups, you’ll know that AI infrastructure startup Honu is the startup profile that has received the most attention here since we published it back in December last year.

But how has Honu progressed since then? I caught up with founder Imad Riachi about:

• Navigating the noise and hype of the AI startup market

• The state of AI agent tech

• The problem of paper-thin defensibility for most AI startups

Read on to see what he had to say…

With this edition, we continue our tradition of resting our regular diet of startup profiles in August. But fear not, the new startups will return at the beginning of September.

– Martin SFP Bryant

Intercom for Startups

Join Intercom’s Early Stage Program to receive a 90% discount.

Get a direct line to your customers. Try the only complete AI-first customer service solution.

Apply now

How Honu wants to cut through the noise in a jaded AI startup market

When we first met Honu, they were building a platform of infrastructure to power autonomous A.I. agents that can help businesses make better decisions (read our profile here).

While it uses AI, it’s not a purely large language model (LLM) based technology. This means Honu is an AI startup that falls outside the current trend for building on top of tech from the likes of OpenAI.

This presents a challenge for Honu in attracting investors who are jaded by seemingly promising startups that end up having zero defensibility.

But it also provides them with an opportunity to enhance underlying AI tech. They just need to convince the world that the agents they enable are smarter and more capable than can be achieved via other means.

Honu’s founder and CEO, Imad Riachi, had plenty to say on this topic when I checked in with him earlier this week.

I’ve edited our conversation for clarity.

MB = Martin SFP Bryant, IR - Imad Riachi

MB: How have things have moved on for Honu since we last spoke?

IR: We’ve built the platform and we’re testing it with a couple of small businesses. We’re an ecosystem play. We're creating something that glues the ecosystem together. It's like smart connective tissue that adds value to everyone.

We've signed an MOU [memorandum of understanding] with a platform operating in the ecommerce space for them to become a distribution partner. And we’ve got a second one coming in, who are very interested in taking our solution as a plug-in agent that they can distribute for their own applications.

We've got a handful of agents that we've created that we know add value. We're putting them in front of users, and then we're going to be distributing the solution to a couple of platforms that will allow us to reach hundreds, if not a couple of thousand, of small businesses.

And then we will open it up to developers of AI agents, and service developers who are looking to ‘agentify’ their services.

We've got a clear plan, we've got working technology, we’ve got two patents filed, and we've got partners who are excited. We just need to grow the team and execute on the vision now.

MB: How are you navigating the noise and hype around AI as you work to get noticed?

IR: We came out talking about something that was very ambitious, and got people very interested. We were saying that if you want to get to full autonomy, it's not necessarily more compute and more data.

We’re saying that as counterintuitive as it seems, the thing that's going to be the biggest leap in AI is actually going to be a software layer where you rethink what your AI stack looks like and introduce a layer that acts like a nervous system.

That’s something that people are not used to, which is both exciting and sometimes frustrating.

It’s exciting because people want to know more and they think it’s interesting. But it’s frustrating because there’s so much noise in the ecosystem. We say we’re building a ‘cognitive layer’ and people say ‘a lot of people are doing that’, and I’m like ‘no!’

I wrote a letter to the FT last week about this: for all the AI, where's the intelligence in the decision making?

There are so many claims in AI that are ungrounded. And we’re building something that really is differentiated, but we're competing for attention. There are so many inflated claims out there that people become desensitised.

MB: Last year, OpenAI got people excited about the potential of smart AI agents based on LLMs, which can perform tasks on the internet on behalf of users.

Your tech is designed to enable this in a different way that could be more effective, but things have largely gone quiet on the agents side of things when it comes the the foundational models. What is the current state of that field of AI?

IR: I think some people have mistaken the eloquence of the output of LLMs for expertise. And I think we're going to go through a slight correction phase now.

A huge amount of budget in bigger companies has gone into building proofs of concept and pilots for AI. This is me being very speculative, but I expect a non-negligible part of this budget will disappear next year.

I think you will see a lot of the use cases that go beyond the suff that LLMs are really good at going away.

LLMs are really good at knowledge retrieval and specific workflows, and people will double down on that, but people haven’t yet had a reality check about what LLMs can do.

I think there will be a consolidation into some of the use cases that have shown to be reliable and a bit more grounding when it comes to whether something’s missing if we want to take things to the next level.

My hope is that people can take a step back and say ‘instead of us keeping going in that LLM-centric direction when it comes to decision making, is there something else that can complement it?’

OpenAI, Anthropic, all of these guys are brilliant, but what if you can plug in something that’s a bit like a digital nervous system? Then suddenly you’ve got more understanding of the principles of business… you can do scenario analysis, you can simulate, you can understand.

I just hope that the adjustment will not create so much skepticism that we would need another AI cycle for people to get over it. A lot of very exciting things can happen; we just need to be slightly more grounded.

MB: What do you think about the state of the AI investment space right now?

IR: There’s a lot of money going into foundational models like Anthropic, Grok [Elon Musk’s X.ai], and Mistral. You can question what how good the results are, but at the very least they're creating the demand [for AI].

But then there’s another bucket focused on results. So you’ve got wrappers around LLMs, and investors are focused on getting them up to millions of users.

But there's so much non-defensibility in the space that what's going up will not only end up saturated, but will actually go down the other way because it's very easy to replicate. And it’s so non-differentiable that whether it’s Anthropic or Mistral, they’ll say ‘this is so easy to do, we’ll make it available ourselves’.

But there’s a space between the foundational models and the wrappers, and that’s where we are with Honu. It’s a platform.

MB: I remember discussing this with a VC on LinkedIn last year, about how little defensibility there is with most AI startups.

A bunch of entrepreneurs disagreed. They said building on top of LLMs is just like building on top of AWS, where the foundational technology allows new businesses to thrive. But in most cases it’s really not the case with LLMs.

IR: The big difference comes when you look at the dynamics of the ecosystem. AWS never had a problem with monetisation like the big AI companies do.

And the difference between the underlying technology and these wrappers is minimal, so if there's anything there that needs to be grabbed I can guarantee you that we will end up being grabbed by those AI companies.

Back at the start of September

PreSeed Now takes its annual summer break for a couple of weeks now, but it’s not really a break as we’re busy chatting to startups to profile from the start of September onwards.

Know a startup we should feature? Drop me a line. We feature super early-stage B2B and deep tech startups based in the UK, anywhere from inception to seed stage.