A few months ago, after we had just closed our best-ever quarter, we decided to do something bold: start from scratch on a clean-sheet version of Hex, the product we had been building for 6 years and used by over 2,500 customers globally.
This was a radical thing to do, but after 4 years on the frontier of bringing AI to data, we knew it was the right one. Today we’re introducing that new version — and announcing a fundraising to bring it to the world.
Let me tell you why, and where we see all of this going.
“What would we build if we were starting from scratch?”… and then doing it
This was the question we asked ourselves earlier this year. See, we were founded in 2019, making us a pre-AI company (ew!).
Now, I also think we had done an amazing job bringing AI into our product over those years. We had a text-to-SQL prototype using GPT-3 before ChatGPT. We launched our Magic features in February 2023. We built the first really good data agent last year, and earlier this year we changed the game again with generative apps. A lot to be proud of.
But fundamentally, we were trying to integrate AI into a product that already existed. Our data model, architecture, and core assumptions around what users were doing in our product were all baked in 2020 to 2021 — we were far from “agent native”, no matter how much we liked saying it.¹
And this was leading to some major issues: our agent harness was becoming bloated with hundreds of tools, which were required to teach agents about how to use Hex. External agent workflows felt clunky, because they were having to use Hex through a keyhole via MCP versus being able to work with something more native. And our compute engine wasn’t built to support the high-speed, high-concurrency, iterative workflows agents seem to like.
In parallel, our customers were experiencing new challenges: primarily a data sprawl from analytics work happening in a lot of other places. Folks are using Cursor, Codex, Cognition, Claude, and custom agents, "vibe coding" dashboards and reports, and sharing the results as HTML files. They’re working around their legacy BI tools, and working without permissions, provenance, context, or observability. It’s a mess, but the previous version of Hex couldn’t solve this in the way we wanted.
Altogether, this added to a growing sense that – despite growing quickly and having a lot of happy customers – we were headed for a point of diminishing returns with this approach.
As we got together and talked about this, we formed of an idea of what we would build if we were starting from scratch today: a platform rooted in code — with all of its infinite flexibility and agent-nativity — but with the context, controls, guidance, and guardrails to make it trusted, no matter where it’s happening.
And so that’s what we built.
Introducing Hex 4.0
Hex 4.0 is truly a from-scratch rewrite for the era of agents².
The core assumption of 4.0 is that everything is code — and that teams need context, controls, and compounding to make it trusted and scalable. It leans into the idea that data work is happening in a lot of places, and that "plain code" (not a proprietary YAML format) is the way to make that make sense.
We have a big, long blog post detailing all of this here, I recommend you read it! Maybe a couple of times, because it's interesting, and we suspect you’re going to be hearing from a lot of other companies who are going to realize that they need to think from first principles too.
And this is just the foundation for a lot of other things we have in mind; in fact, one of the reasons we went so hard on this is because there are things we want to build we couldn't imagine doing on the old architecture — but are trivial on the new one.
We can't wait to show you!
Some money stuff: we raised another $100m
As this all came into focus, we realized more money would help us go faster – to scale our team, expand go-to-market, and continue to make great swag. The only problem was that in the first part of the year, investors thought all software was going to die and there would only be two companies left.³
Shailendra and the team at Peak XV, however, didn't think that: in fact, they were pretty sure that model-independent platforms were going to have a bright future ahead. And they were not scared about backing us taking a slightly crazy, first-principles approach to go build it.
So I’m excited to announce that we’ve closed a $100 million financing round at a $1.2B valuation led by PeakXV and 137 Ventures, with participation from Amplify, Avra, a16z, Next Legacy, Redpoint, Sequoia, and Zero Prime.
This gives us a bunch more runway to keep building this thing, and take more bold bets into the future.
The end of BI, the beginning of something new
I'm throwing some serious shade at the "SaaS apocalypse"-pilled investors above, but I'll give them some credit: it is true that a lot of software categories will effectively die over the next couple of years. But I also believe many of them will combine and be reborn into new shapes.
This is our basic outlook on BI: the old paradigm is over, and no amount of bolting a chatbot sidebar onto the legacy platforms is going to change that. Agents don’t want to use the bespoke UI you wrote, and they don’t care about your esoteric data model.
But these currents are also opening up fantastic new opportunities! The basic promise of BI is finally here: everyone can ask questions! Everyone can build! But now it's chaotic, and all the trust is gone. We see this as the core trade-off to break, and we're kind of obsessed with it.
We believe the next chapter of data work is rooted in code, with the context and controls needed to establish trust, governance, and guardrails. And we’re so excited to show you what we have coming next.
¹ I'm not going to call out anyone by name, but I do think it's funny how many companies in our space that have been around for 5 years or more now call themselves "AI native", despite that literally being the antithesis of that phrase. I mean, I get why they're doing it, but unless you have literally rebuilt your product in the last 24 months, I'm not sure you can credibly claim it...
² And, of course, was greatly accelerated by using agents to write all of the code, guided by the decades of combined expertise from our teams in building scaled data platforms.
³ I'm joking, but not really? You should have sat in some of the pitch meetings we had…