Ben Haynes

Founder & CEO of @Directus

[Ben Haynes @benhaynes]

I just wanted to take a moment to thank all of the 15k+ @directus stargazers! GitHub stars are our most meaningful metric, and we sincerely appreciate everyone's support.✨🥰✨

https://t.co/GiMmmEoUaT

#opensource #sql #database #api #nocode #javascript #cloud #100daysofcode

[Ben Haynes @benhaynes]

There's a Cursor story I keep thinking about from a few months ago.

They deleted their CMS and vibe coded a replacement… over a weekend, I think?

For them, it made total sense. Tiny team. All engineers. No marketing ops, no compliance review, no 15 people editing the same record on a Tuesday afternoon. So of course it worked (though I'd be curious to hear an update)

My only issue was the lesson people seemed to take away from it.

It wasn't "Cursor's context is unique," it was "we can vibe code everything."
And that's where I start getting a little nervous, haha.

#####

Here's what I truly, deeply believe (+ video of me actually talking about this at our recent Leap Week event):

You can vibe code a UI in a day.
Forms, dashboards, components, workflows… all of it.

But you CANNOT vibe code trust.

* You can't vibe code permissions that hold when HR edits an employee record next to marketing editing a blog post.

* You can't vibe code data integrity when five teams need the same record to mean the same thing.

* You can't vibe code uptime when traffic spikes and everything falls over.

The UI is the part you demo. The backend is the part you live with.

#####

So when I see another vibe-coded app shared as production-ready… I'm not dismissing it. It's genuinely impressive. But a clickable prototype isn't software.

Software is what happens when the prototype meets real users, real data, real edge cases, real stakes… and still holds up.
The easier the front end gets, the more exposed the backend becomes.

#####

Ok… so maybe "you cannot vibe code trust" is a little dramatic.

But trust isn't a component you can just import.

[Ben Haynes @benhaynes]

@rijkvanzanten What type size??

[Ben Haynes @benhaynes]

A few weeks ago I opened Leap Week with a keynote called “The UI Era Is Over.”

A few people pushed back on that title… which I love.

It’s a bit sensational… and if I’m being honest, there’s a lot of room for interpretation. I’ve been thinking about it quite a bit since.

What I believe is this:

For a long time, we’ve mistaken the interface for the product.

You click buttons. You see a dashboard. It looks innovative, polished, intentional… “done.”
That’s what people point to and say, this is the software.

But that’s never really been the whole story.

The UI is just the surface.
The part that’s easy to show.

What actually makes software work lives underneath it… and like the old “icebergs are 90% below water” visual, that backend is bigger, harder, and way more complex than most people realize.

Engineers have always known this.
Now everyone’s starting to feel it.

Because AI just changed the game.

You can spin up a decent UI in minutes now.
Forms, dashboards, components, UX/UI… all of it.
It looks real. It feels real.

And that’s incredible.

But it’s also a little misleading.

It creates this illusion that building software got easy.

It didn’t.

The frontend got easier.
The backend is just as hard as ever.

And when everything looks “good”… looking good stops mattering (unless you still somehow mess it up, haha).

We now have more tools, more builders, more front ends than ever before.
Which means the front end isn’t where you win anymore.

So if the UI is becoming commoditized… where does the value go?

Down. The. Stack.

Into the layers people don’t see… but absolutely experience.

The logic. APIs. Permissions. Data pipelines. Uptime. Data integrity.

The stuff that doesn’t show up in a screenshot… but shows up immediately when things break.

You don’t notice great infrastructure until it fails.

When your app goes down.
When permissions break (or get exploited).
When data disappears.
When a spike in traffic takes everything offline.

That’s where trust lives.

And that’s the shift I was trying to point at.

The frontend still matters. Amazing UX/UI is now table stakes… which raises the bar.

But it’s no longer the thing that pushes you above the noise.

The thing that differentiates you is whether your system actually holds up under load… at scale.
Whether it can be trusted.

Ok… maybe the UI era isn’t “over.”

But it’s definitely not where the center of gravity is anymore.

That’s moved underneath. Deeper. Down the stack.

[John Daniels @iamjohndaniels]

The fastest way to go from "we need a backend" to "we have a backend" is to not build one from scratch.

Directus gives you the 80% that every project needs so your team can focus on the 20% that actually matters.

[@levelsio @levelsio]

Okay honestly this makes vibe coding into production very dangerous, you guys were all right

I think what I'll do is cut off all access to DBs and run it as a user with almost no privileges

[Ben Haynes @benhaynes]

@bryantgillespie @mattm1nor @iamjohndaniels @directus @openhome Bahhahaha

[Bryant Gillespie @bryantgillespie]

I've been a locked into a vibe competition with my colleagues @mattm1nor and @iamjohndaniels. Well gents - hold my 🍺 .

I've now got a sarcastic voice assistant that can do real work inside my @directus backend and roast my colleagues at the same time.

Thanks @openhome 🙏

[Alvaro アルバロ @alvarosabu]

It's finally here!

My new portfolio is available now. Made with @tresjs_dev and @nuxt_js as result of my study of @threejs shaders over the last couple of years.

The home page contains 3 different variations, can you find them all?

More about the project on thread 👇 (01/04)

[Ben Haynes @benhaynes]

@staysaasy Haha, sounds about right!

Have you considered using a data backend platform for this AI-assist frontend? That might help give some proper access to your data model... and provide a solid foundation to vibe off of.

[Marc Backes @marcba]

Son of a bitch, it worked! 🥳

- Bun backend with Vue-powered reactivity
- Controlling a synchronized routine 1000 phone screens
- Sync corrected for system time offset with NTP

It worked better than I would have ever imagined. Thanks for everything supporting me on this talk ❤️

[Rijk van Zanten @rijkvanzanten]

@AnthropicAI hey our thousands of $ annual team plan suddenly got canceled for "suspicious signals", but I have no clue what could've caused that and have no way to resolve on my end. This is very impactful to my business, how can we resolve this?

[Ben Haynes @benhaynes]

Two years ago we ran Leap Week. Then life happened, and we didn't.

We're bringing it back. March 24–26.

The reason now feels right: something is shifting in how software gets built. The interface used to be the differentiator. When anyone can spin up a frontend in minutes, it stops being the moat. The value doesn't disappear... it moves.

Where it moves to is exactly what we want to figure out together.

Nine sessions. Partners we'll be announcing over the next two weeks. And the kind of conversations I've been wanting to have with our community about everything happening right now.

If you're a builder (or are becoming one), this one's for you.

Register → https://t.co/vtcOjMmQw0

[Directus 🐰 @directus]

"The First Episode" https://t.co/3Qp3Hm61mW

[Massimo @Rainmaker1973]

Programmers were asked to make the worst volume control for a contest
https://t.co/DcwMiDAGm2

[Ben Haynes @benhaynes]

@iamjohndaniels @CleanShot 100%... it's so core that I forget it's even there, haha

[Ben Haynes @benhaynes]

Snowed in here in Connecticut today. Good day to reflect.

Last week we had a quarterly board meeting… and it had me thinking about the people around that virtual table.

My exec team. My chairman (who started as my first and most trusted advisor). Board members from our investors. A few observers. When you zoom out, it’s kind of surreal.

Bootstrapping is hard. Fundraising is harder... especially right now, when VCs are comparing everything to AI companies posting absurd year-one growth numbers… or writing big checks into pre-revenue, pre-PMF AI dreams (hallucinations??).

And even if you raise successfully, money is just money. It only gets you so far.
The real variable is who you partner with.

VCs come with capital... but also ops teams, recruiting help, experienced pattern recognition, and networks that can open doors you simply can’t on your own. Choosing a firm isn’t just about terms/valuation. It’s about alignment.

Are they active or passive? Are they the kind of partner you’re nervous to bring hard news to… or grateful to have helping your through it? Do they truly understand your model and ethos (eg: OSS, dev tooling, exit ambitions)? Have they actually lived it as operators?

If they’re leading your round, they’ll likely take a board seat. That means real influence over your project’s future.

Are you choosing a firm’s logo… or their specific sponsor? What happens if your sponsor leaves the firm? Do they want growth at all costs... or responsible growth?

More importantly… which do you want?

A few years ago, I knew literally nothing about any of this. No MBA. No playbook. No ChatGPT. Just a design degree, some light technical skills, and an equally capable, excited, and motivated co-founder (hey Rijk!! 👋).

YouTube got me started, haha... but my investors helped me actually grow into my role as CEO.

That’s why it matters who you team up with (in business and in life).

It can feel like you should jump at the first term sheet. Maybe sometimes you should... survival matters.

But these are long-term relationships. Do your diligence. Ask them for references. Talk to their founders (not just the successful ones).

Five years after converting our OSS project into a DE C-Corp and raising a $1.5M SAFE (microscopic by today’s standards)… I still feel great about those early choices.

Yesterday’s meeting was long. Some topics were more controversial. But it was honest dialogue around things that mattered, it was about alignment, and it was respectful and exciting (from what I’ve heard, this isn’t the norm). And that’s not luck.

I’m extremely grateful for Handshake Ventures, True Ventures, F-Prime, Eight Roads, and PWV (Preston-Werner Ventures). Not just for their initial trust, but for their ongoing support in this shared vision of Directus.

Capital helps you grow. The right partners help you build something that actually matters.

[Marc Backes @marcba]

SO EXCITED. To show you what we’re working on at @directus 🐰

[Ben Haynes @benhaynes]

All these AI-generated apps are awesome… right up until they become important. 🫣

They look great. They work. They solve a real problem.

Then they go into production or maybe just need a bit more scale, and they're handed off to engineering... who quietly starts over.

An AI demo is the easy part. Production... not so much.

For the last 20 years, software has been judged by its interface… innovative UIs, polished design systems, and endless front-end frameworks. AI is starting to flip that upside down.

Today, a decent interface can be generated in minutes. Tools like Lovable and Replit are turning marketers, founders, and operators into builders. We're moving from a society of software users to a society of software creators. 🙌

That's exciting. It's also a little chaotic.

At the Enterprise level, most of these AI-generated apps are closer to proofs of concept than production systems. They don't account for permissions, audit trails, structured APIs, automation, or long-term data integrity. They weren't designed for scale, governance, or reliability.

So when they become mission-critical… they get rebuilt.

And as the number of front-ends explodes, another problem appears: you can't have a one-to-one backend for every AI-generated interface. Ugh, no thank you… my browser couldn't handle the tabs.

What you need instead are durable backends that can power many front-ends at once (I thought we already learned this when sites went "headless" haha)… stable systems of record that outlive any single interface.

Another option is removing the backend entirely for simple cases. Cursor recently shared how they replaced their AI-generated website's CMS with raw code and markdown. That works when your whole company is made of highly technical devs and the system isn't "mission-critical"... but most orgs don't live in that world.

And while AI can generate interfaces today, building a truly production-ready backend with AI alone isn't really an option yet. For now, the data engine still needs consistency, carefully considered structure, future-looking specs, and proven systems.

In the next era of software, interfaces may change every day. The front-end is becoming disposable.

The backend is becoming the product. 🧱

[Directus 🐰 @directus]

Every team needs backup.
Your own AI Assistant in Directus. Coming soon!

[Rijk van Zanten @rijkvanzanten]

Very excited to be back at this blast of a conference. See you there? 😃 @vuejsamsterdam