Racing to the Obvious

I debated if I'd send this post out as part of my newsletter as it gets a bit ranty, before eventualy decided to send it. That said, if you want to receive all my drivel you can follow my RSS feed or my Mastodon. Both are totally real services that still exist, I promise.^[1]

Recently, I've been noticing a pattern as startups struggle to jump on the AI train. I've started to internally refer to that pattern of companies as "racing to the obvious." They are all building or pivoting to some of the most low-hangingest of low-hanging fruit that they can still slap an "AI" label on. The copy talks of magical features and gaining back hours in your day.

It's possible the average person might not notice so many companies are chasing the same couple of ideas, almost all of which are really just light wallpaper over top of OpenAI or a handful of other APIs (say it with me, if most of your business is a couple prompts, you don't have a business).

Perhaps if you just see one and aren't already using ChatGPT Pro, Copilot Pro, or whatever Google is making these days, you might think it's a novel idea. If you're perusing Product Hunt, looking at App Sumo, or paying attention to new company announcements though, you start to have a lot of déjà vu.

The Lifecycle of Product

Where I make product lifecycle sound a lot like playing Spore^tm.

I've been a user of krisp.ai for four years (since 2020, which feels longer ago now than I'd like). Krisp started out as a tool to make noisy and cruddy audio from headphones (or, horror of horrors, using the built-in laptop mic) sound acceptable-to-good to others. I often had a knack for working from coffee shops back then, and AI noise reduction was my jam.

In April 2020, Nvidia launched RTX Voice and started to threaten the market share of Krisp. Also in 2020, Krisp signed a deal to integrate their technology into Discord. I wonder if at that point they saw the writing on the wall for noise reduction software, as Zoom built this as a feature later in 2020 as did Microsoft Teams in 2022. The ultimate end of this as a unique feature was probably when MacOS Ventura added noise reduction for all audio devices in 2022.

So this is a bit of a cautionary tale, right? It aligns well with my own view of software products (which is also in alignment with Ben Thompson's aggregation theory). Namely, a software product is an unstable state of matter and cannot exist independently for long. It will meet one of six fates:

Featurization

Krisp is an example of the featurization of a product. There wasn't a use for Krisp on it's own, as you always needed to use it with another product. Those other products then added the "krisp" feature directly to their product to add differentiation, and eventually they all had it. Suddenly, it was universal, and there wasn't really a compelling use case for someone to start using Krisp unless they were using an app that didn't have the feature, and using a platform that didn't have it built in to their OS or Video Card drivers. In startup speak, their TAM (Total Addressable Market) went from everyone who uses online meetings to a few niche cases of online meetings. That's not good.

Krisp, however, aren't taking becoming featurized lying down, and so they've attempted to expand their product's functionality as well to avoid becoming replicated entirely. They've initially added recording, which is an obvious value add, but also keeps them very much in the territory of things that can be featurized (and have been) by the products they rely on to function.

Their most recent work to attain/retain relevancy has been to go deeper into the AI side of their name. That is including transcription with a speech to text engine, and then "meeting notes" with an LLM model to provide a summary and to attempt to detect a list of action items after the call completes. I've found the call summaries wildly hit-or-miss, and I don't think my experience is unique as Krisp's own sales team discloses that they use Gong AI for call recording and meeting notes* in their data processor disclosure.

Again, these are features that are ripe for featurization, and that featurization has already begun. Aside from the sales platform they themselves use (Gong) having this functionality built in, both Zoom and Microsoft Teams offer on-call transcripts and AI summaries as product offerings.

This push and pull of product development is nothing new; and the onus will be on Krisp to either make strides to add new features that won't duplicate ones now built into other products, or to pivot into products that won't be as easily disrupted.^[2] Krisp last raised in 2021, and I have to imagine they (and their investors) are nervously looking for their next step.^[3]

Diving Into The Clown Car

Missing The Boat As A Service (Mt. BaaS)

You might have thought this article was about Krisp, but no, Krisp was here to illuminate the point that they built functionality and then had other platforms adopted that functionality. They're inventors in a tough spot innovating in a fast moving space where a sole champion hasn't been chosen yet. With Google Meet, Zoom, Teams, GoToMeeting, and Slack all competing with the same general product anything Krisp makes will get featurized by one of them in an attempt to eke out more market share.

No, this article is about companies who keep trying to make the same product many other companies are also already making with the thinest possible layer of innovation. Diving into an already crowded space, either through complete lack of awareness or a belief that they're special matched only by a 2nd grader who gets the part of Jesus at a Sunday School play.

Limitless

Limitless is one of the companies bringing a hardware AI device into the world. After Humane AI and the Rabbit R1 cleared a path in the market by lowering people's expectations underground to the point where "actually looks useful" could clear it. Limitless' "Pendant" has come out looking like an AI device that actually has a use. Ultimately the Humane AI and Rabbit R1 didn't have software that worked as described and didn't have hardware that did anything your phone couldn't have done better.

Limitless paves a different path by doing something your phone can't do, record audio. It's weird that phones never came with any sort of voice memo tool, or any apps that can record and transcribe meetings. But, they don't. Notes Tom's Guide, "Once turned on, the Limitless Pendant records everything you hear and then uses AI to transcribe conversations so that you can remember them later. However, Limitless takes things a step further by suggesting action items for your to-do list based on what was discussed during a meeting."

Limitless is joining no shortage of apps (and physical things, too, in this case) that are productizing a very thin UI on top of things you can already do, where players in the market already exist in droves, and where platforms themselves either already have or are adding these features. How is limitless going to stay relevant? Who can say.

Oh, and until their mic broach arrives you can install an app that records your system speakers and mic and transcribes your meetings for you. Like krisp, just sans the audio enhancement functionality. Limitless AI costs $19/mth if you spring for an annual subscription. For comparison Microsoft Copilot which can do the same thing for meetings is $20/mth.

Granola

Granola "takes your raw meeting notes and makes them awesome." While you're in a meeting you can make some rough notes, and when it's done it uses speech to text and an LLM to enhance your notes with more details from the meeting, tied to a transcript.

It's another app you install that listens to your system audio and microphone, uses an API to transcribe, and uses an LLM to summarize and generate text. According to their website they use ChatGPT 3o for at least some of that. Granola checks in a little pricier than Krisp at $10/mth. You can also chat with your meeting notes.

Otter AI

Otter AI has been around for many years and was an OG app for live transcriptions. As transcription technology became more common though they've branched out. What's the next logical step from AI transcription? You'll win no prize for guessing AI meeting transcription, meetings notes, and some sort of chat bot experience. Otter AI is also $10/mth, limited to 20 hours per month of audio, or $20/mth for 100 hours per month.

timeOS

I'm just going to let their marketing copy do the talking on this one.

tl;dv

Does meeting transcription. The unique element for them is they'll also do reports across multiple meetings, and they seem to be building more Gong-like sales tools. $19/mth.

OK, so the space is crowded...

Oh, I'm not even close to done here. I'm sure there's a hundred more of these and another launching every week. Fireflies.ai ($10-19/mth), Avoma ($19-79/mth), Summie ($10-19/mth), Dovetail ($29/mth), Laxis ($15.99-29.99/mth), Loopin ($12-18/mth), Fellow.app ($10+/mth), Tactiq ($12-20/mth), Meetgeek ($15-29), Collato ($10/mth), Briefly ($15/mth), GoodMeetings (?).

Many of these companies will have a slightly different frame (collaboration, sales, 1:1s, templates, slightly different UIs) but overall are meeting summarization and LLM engines tossing a transcript and a prompt into LLM model.

Chasing Fads

Turns out "Magic" alone isn't enough.

So what's my point here (aside from there are seemingly unlimited AI meeting assistants all at risk of just becoming built-in features tomorrow)? I see these companies falling into two camps, those who had existing products before the AI wave and wanted to find a fast path to get "AI" into their product, and those that started after the AI wave and mostly wanted to get an AI product in the market quickly.

Let's talk about an unrelated topic for moment.

There were these insulated mugs.

They were building popularity for their ability to keep things cold for a really long time (because insulation). Then someone crashed their truck, and, even though it had been on fire, the ice in the mug stayed ice!

Insulation in action, it was incredible! People lost their minds that insulation insulates things. Then the company that made the insulated mug bought the person (whose social media post was blowing up) a new truck. It blew up even more (the mug craze, not the truck).

It was a snowball of mugs! Everyone wanted one, and they sold out, every one. Prices jacked up; suddenly, people weren't buying these mugs because they were cost effective for the value, but because of brand-they-know from the truck flambé.

Then people discovered the mugs contained lead (and remembered that lead was pretty bad, they were quite sure) and the fad faded fast.

Keep that in the back of your mind. Let's return to the AI meeting assistants.

I suspect the first company to make meetings notes and action items and apply LLMs to transcripts was pretty excited to share what they'd done and could be pretty proud in applying new tech to make a better product quickly. The next few likely also wanted to offer a similar product and compete on quality or price.

All of them would have had to know this was a feature the product-their-products-depend-on-to-work (the meeting software) would eventually integrate, such that it would be a short lived edge. For those early movers, however, I see the value proposition. There's an angle there.

The question I've been bothered by has been, "Why did the 10th company build an AI meeting assistant tool?"

At this point, the main hope for the existing players in the space is to be able to create new experiences, or to be able to build out an ecosystem of related tools (much like Gong has done and tl;dv is trying to do). Why would someone be pivoting their company (like Rewind AI becoming Limitless) or building a new "AI Meeting Assistant" app now? I think the answer lies in getting on the AI train.

After the pandemic tightening of VC funding, one of the few areas VCs have continued to be willing to invest is if you're doing something in "AI". Companies are reacting to induced demand, not from their customers, but from their investors who will join every call with the question "how are you applying AI in your company?"^[4]

This is detaching companies from building products that suit the market and their customers needs, and instead to suit investors who want more AI companies in their portfolio and don't want to look left out when they see what other VC firms have in their portfolio.

Sadly for those, all the off-the-shelf API-based ML tools we have in the world still hallucinate, still don't have any sort of true intelligence, and mostly have turned "fake it 'till you make it" into an expensive service. ML tools will get better, and eventually AI will start to have intelligence, but we're not close to that day. I feel like we're in a similar state with AGI as we are with self-driving cars: the last 20% is going to take decades.

What results is you have a wave of companies all rushing in to implement the same core use cases: speech to text, summarization, and re-writing. It's how you end up with a never ending supply of AI Meeting Assistants, Chatbots, and image generators.

We're racing to the obvious. Just like with the mugs: who knew that insulation could insulate? Though if anything in this is AI's metaphorical "lead content," it's the aforementioned hallucinations and whatever nightmare-chimera of copyright and privacy issues we'll eventually find ourselves battling.

For now, though, that mug is pretty neat. Just don't overspend on it, or get too fixated on trying to corner the mug market. It's pretty saturated already. At least try to sell me a novel new mug holder, or something special to put in the mug, or a proper mug-catering service that can provide mug at scale.

Or, maybe, move a little beyond the mug first, and then we can take stock of where we're at again.

If you found this article interesting, send it to a friend! Word of mouth is the only way folks find my blog.