Firedrop: A website builder that does everything for you | Hacker News Reader

Firedrop: A website builder that does everything for you

firedrop.ai

83 pointsspking60 comments

This is exactly like The Grid, even down to their landing page marketing. The grid was such an obvious fuck-up I wrote an article about it several months ago [0]. I have never seen such a difference between hype ("AI Will design your web page - Never hire an engineer again!") and reality ("Everything looks like shit and takes forever, get your refund fast before we go broke").

[0] https://medium.com/@seibelj/the-grid-over-promise-under-deli...

I thought it was The Grid.

Hey guys, Marc from Firedrop here. I completely understand the negative comments here about The Grid. I was super excited when The Grid came out and paid up the $90 like a lot of people, only to be hugely disappointed when I actually used it. They're pioneers in this space and they've done a lot of really interesting and cool work... but they've misunderstood their users hugely and made too many assumptions about them regarding design decisions. We're taking a different approach where our platform listens and learns more closely from the users rather than whatever broad abstract data set they're using. It'll lead to less "out there" designs that people will be happy with.

Kudos for going for it!

What, in particular, would you say you're doing better than The Grid? Which design assumptions are you not doing, and how is your approach different in terms of learning and listening to the users?

Thanks for the kudos! Hopefully we can deliver where The Grid couldn't. We actually paid close attention to what they did and think we've learned a few things.

The main difference is that we are spending more time learning what the user likes first before offering design options. Options being a key word here, because The Grid doesn't really have much of a feedback system and even their latest V2 alpha resorts largely to guesswork, whereas our system will give you choices and ask questions like "Do you like this?" at the appropriate points.

The experience we're going for is the one you get from working with a human web designer: at first it's a two-way discovery process and then, over time, the web designer gets to know and understand you and therefore becomes better at nailing what you want straight away. Simultaneously, the platform will learn from anonymised behaviours from its other users, mixed with analytics monitoring, to discover recommendations that it can pass on to you as a user. For example, imagine you're a restaurant owner and you have a button that links to your booking engine which is labelled "Reserve a table". Our platform learns, from other restaurant sites on the network, that "Book a table" has a higher conversion rate, so it can pass that recommendation automatically on to your site. That kind of thing.

I've never used The Grid or anything like this, so I'm curious: How much of your landing page is designed using Firedrop itself? This landing page seems like it's the perfect use case, so it seems like a good opportunity for dogfooding.

Nice to see another team go at this problem.

Firedrop's marketing using the general term "AI" doesn't give much confidence. As a startup, you probably want to keep your cards close to your chest, but the current stance is too closed and not confident enough. Google published page rank algo in 1999. Even TheGrid published GSS. Anything worthwhile like that to point out?

Designing websites with the help of AI methods is hard. thegrid.io is competing in this space, and I'm not impressed with what they have achieved so far. Initially they stated that they would launch some time in the spring last year. It's been over a year now, and they still haven't given all the beta users access. Yesterday, The Grid featured a website [0] created with their tool. That website takes around 25 seconds to load, even with my speedy internet connection. Also, I wasn't impressed by the menu, which was merely a hamburger button. I think they need to focus more on UX and faster loading times and less on fancy image processing and color matching algorithms. I wish Firedrop luck going forward, and hope that they learn from The Grid's mistakes.

[0] http://curlqueen.salon/

The approach they took looks to be fundamentally wrong. If they had started with some solid base templates and made some parts changeable depending on the content, they would have had a product that came close to what they promised.

The source code for this page is bad in so many ways..

The ironic thing is despite apparently not using solid base templates, the end result looks like it was knocked up by a teenager (ab)using a generic modern WordPress template anyway... until you check View Source and burst into tears

The worst design issue isn't so much the questionable taste as the fact the "AI" doesn't seem to have figured out that the primary purpose of the website - to book appointments - probably shouldn't be hidden away in a hamburger menu.

Wow. That page loads 41 MB.

Including one 18.9Mb gif and no less than 6 jpgs clocking between 1.9 and 5.9 Mb. Looks like counter-optimization at this stage.

At least you can assume that all of your images will always be full-resolution.

the AI has clearly found the trend with pages over the years and is trying to make the page feel more futuristic

Can you upload a video or a bunch of screenshots of that website? It’s been 6 minutes on my 100/40 connection and it still hasn’t loaded more than the logo.

I’m quite sure at this point a 4K video showing the site would load faster.

EDIT: It has loaded right now. I checked with speedtest.net and fast.com just to check if maybe my connection is at fault, but nope, that site is just horribly slow.

And react.js? For a static site? WTF

Resume Driven Development

Dear god do not view this site on a phone with a data cap.

50 reqs, 41MB transferred (!), full load in 23.06s, DOMContentLoaded: 1.11s.

Insane.

So https://thegrid.io/ ? Without the big names behind it?

I gave "The Grid" a spin myself but haven't really used or updated my site due to lack of time. I browsed the directories of other peoples' sites [0] manually and only 1-2 were "alright" with most of them being rather bad. I'd be disappointed if my expectations for it weren't near rock bottom to begin with.

Due to it being A.I-based the only thing I can really say is "It's been tried before. Best of luck on doing it better."

[0] https://github.com/the-domains

I don't see any reason to believe that just because the idea failed in implementation once that it will forever. In all likelihood, if we try at the problem long enough someone will eventually figure out algorithms that work fairly well.

Perhaps "AI design" competitions or benchmarks would help this line of work. But people trying multiple times can only be a good thing, even if the early attempts are not great.

>Due to it being A.I-based the only thing I can really say is "It's been tried before. Best of luck on doing it better."

I'm sincere in saying that. But unless you somehow draw up some marketing hype - it becomes increasingly unlikely to receive funding for an idea that's failed in the past unless you have some tangible evidence about how you're avoiding running into the same problems.

If (/when) they have an actual product we'll see if they managed to do it better. But there have already been enough failed attempts (TheGrid wasn't the first or only one since) to not be hyped because "AI codes my site".

So with nothing else to go off of other than "This is attempt #x by some new people (or person)" I can't say I'm excited or have high expectations. I'd be lying.

The logo is really, really close to that of Lightspeed, https://www.lightspeedhq.com/

Lightspeed was probably in the training set ;)

No the logos did not look so close to me as you claimed.

I'm really not seeing how you can just take content from a client and algorithmically design a website that will make them happy. It's hard enough doing it manually with lots of feedback and iterations. This sounds well beyond what current AI can deal with from first impressions but I'm willing to be convinced.

It's a case of sensible compromise and balance between offering suggestions, taking feedback in frictionless ways, and learning. I've learned from running digital agencies over the years that a lot of the processes that you go through in a web project can be distilled down to heuristic choices and that's what we started with. The rest us refinement through machine learning techniques.

If this works, I think the technology is exactly what's needed for the tool to become incredibly popular... but the marketing is slightly off.

The message is that the AI will do all the work for you, make all of the choices for you. But people like to design things! Not the precise work of arranging layouts, colors palettes, images, etc., probably, maybe not building things from scratch when they don't know the tricks needed to achieve that intangible feeling of professionalism - and certainly not the coding that most people have no idea how to do. But being involved at the last step, things like being given a choice of layouts and asked to render an aesthetic judgement on which is best, which needs no training* - in my experience, most people love that. It gives a sense of control, personalization, accomplishment. And on the flipside, the feeling that one has no control can be quite negative and frustrating.

But in many cases, choice is something an AI designer can offer as an advantage over traditional methods. After all, if you don't want to shell out the money for a professional custom design, you're typically stuck copying other people's templates with at most minor changes. In theory, an AI can let you make more meaningful permutations to the design to make it feel more personalized. Heck, I think even just the perception that a design is "from scratch" rather than "copied from someone else" would be valuable, even if your "AI" is dumb enough that in reality it amounts to the same thing.

If you do want to shell out the money, well, I don't expect a site builder to be able to compete with the results of that - but it's worth noting that you can spend as much time as you want nitpicking an AI design without worrying about wasting the designer's time.

I can't really tell from the video how many actual customization options the tool will have, but I think it should have as many as possible - all optional, of course, and probably most trivial - and be marketed like "anyone can design a website in a few clicks", not "we'll do it all for you".

*well, experience is probably required to actually pick the best one, but if all the choices are good then it doesn't matter much :)

That's actually spot on. I ran digital agencies for the past few years and it was always important that the client had some kind of input to the process, regardless of how good the design concept was. It's a simple pyschological need for a person to feel some kind of ownership and connection with their site, which is where I really feel The Grid has gone wrong because their designs are so far off what people want or are used to.

We're updating the site in the coming days to demonstrate better how we involve the user, and presenting different layouts is actually one of the ways in which we do it. Firedrop is more of an A.I. web designer than a tool: you interact with it in a much more engaging and human way.

Yeah, I think "collaborating with an AI designer" would be a compelling way to explain it! It's helping you design your site the way you want, not doing it for you.

Same idea as the horrific trainwreck known as The Grid.

I've heard about The Grid and paid attention for a little while with some excitement, however lost track of them.

What do you mean by a "horrific trainwreck"? What happened... nothing?

Took ages to finally get access and when access was finally dished out - the resulting sites from the A.I were quite... terrible. Not 1990's Web Design bad but not much better.

https://www.reddit.com/r/web_design/comments/49djg6/finally_...

See (the links in) this thread: https://www.designernews.co/stories/65265-the-first-gridio-s...

It's quite frequent to see the same ideas repeated over and over by "young" entrepreneurs. I've spent a years in multiple incubators, and they're _always_ a bunch of projects ("a social network for xxxx", "a marketplace for yyy") that looks exactly the same with similar marketing. Sometimes they would even share two extremities of the same open space.

But, it's not that bad. Everybody has time for mistakes. And, maybe they'll even make a bit of money, learn a few things along the way.

I think the idea behind The Grid was good but perhaps the execution was not so great. We're aiming for a better execution of the concept, with more of a focus on what the user actually wants rather than trying to artificially deconstruct the fundamentals of design as a concept (which is where The Grid get it wrong, I think).

"The grid" all over again? I still feel robbed of my $90...

Explain

I'd guess that it's related to https://thegrid.io/

Also

https://www.designernews.co/stories/53246-discussion-is-the-...

http://www.webdesignerdepot.com/2014/12/is-the-grid-a-better...

https://medium.com/@austinwulf/the-grid-io-builds-incredibly...

The grid.io was an AI version of something like squarespace. The premise was you wrote the content and uploaded the images, and it would select the best design for you.

Unfortunately, it didn't work well in practice. This was only discovered a year after tens of thousands of people signed up, when it was finally released.

Big impact for the announcement video. Small impact for the release.

Without seeing any example work on the page it's difficult to tell what's going on, but maybe they're going to

(1) Raise a bunch of VC money

(2) Farm out website design to Fiverr/Bangladesh at cost

(3) Raise some more VC money

...

(?) Maybe plop some stuff into a RNN model and see what comes out

It's Unbounce with a gimmick. Is the gimmick a 10x innovation? Unlikely, since it seems likely the designer would have to go in and do their own edits anyway.

The GIF on their site makes it seem barely more impressive than using a basic 'make this in half an hour in a hackathon' template. But the bit about continuous improvement is interesting:

"Our learning algorithms monitor your visitors' behaviour, perform automatic A/B testing and make continuous incremental improvements to your site in order to improve conversion rates and help your website to grow."

Consider me skeptical.

Continuous improvement through machine learning seems like fodder for a really fun 4/8chan thread

I'm curious about the relevant economics here.

A fully automated system for designing websites could make it easier to design good-looking websites without hiring a designer.

If your system works perfectly, that allows someone who wants a website to save: (1) the transaction cost of hiring a designer, and (2) the cost of the designer's labor (wages). Additionally this person has to "pay" the cost of learning to use the automated system.

Which of factors (1) or (2) is larger? I imagine it's probably (1) -- the hassle of hiring and managing a designer is large relative to the actual cost of labor. Does that sound right?

I wish there was something less fancy, less automated, less AI, but that just displayed content we dumped into it in a more-or-less sane way.

Write anything, upload files, anything, and all will be there in the website somewhere.

Probably not quite what you have in mind, but I use Camlistore like that ... though not for a public site.

I've tried Camlistore in the past, just because of that ideal I have in mind (if I remember well).

TheGrid's cool contribution has been GSS - "grid CSS" - where the style sheet is resolved using a constraint solver.

From there to claiming "AI" is probably the mistaken view of the original AI wave.

Oh god. Not again.

It almost feels like this piece would be a far more compelling product by itself:

I'm fairly "meh" on the main concept of an AI-designed website (though good luck!), but the piece I highlighted above? I'd be way more interested in that, personally.

So much negativity in here. I personally think that A.I. website builders are the future. I haven't used Firedrop, and can't tell too much from the landing page, but anyone exploring the space is worth following I think.

@marc_firedrop Mind if you can share what your training set is?

I can't share the set itself but can tell you a portion of it was made up of scans of open-source website template that we've hand picked based on popularity and quality and then labelled using Crowd Flower. The minimum we are looking to achieve is websites that can compete with anything you might find on Themeforest or Wordpress.

"Is it just me, or do all websites look the same?" This site looks so different that it must have been built by an AI:)

a while back, I interviewed at a company that wanted to use ML (tesorflow was mentioned in passing) to create something similar (they were trying to create Shopify + this).

I was extremely interested in the project, but sadly I was told that my skills didn't align to their new pivoting vision.

Meh

Yup

Yay

(Wait, is enthusiasm for people trying something new bad here? If so, "Meh".)

Don't try to copy whatever you think the accepted response on HN is. Make your own opinions and stick with them.

It was a (failed) joke in reference to a flurry of initial negative/content-free responses. I was trying to be gentle about it, but I'd have done better by being as straightforward as your response. As an HN n00b, I appreciate the feedback.

From a designoper POV, the idea is sound (visual and information design isn't magick) and I wish them all the luck in the world. A good solution will fill a real need.

The comments about "The Grid" are irrelevant, except as (surely known) lessons to the Firedrop creators. Anyone with industry experience understands that attempts will fail, until they don't.

Reply on news.ycombinator.com