YC’s latest batch was indeed “Maybe AI can do…is this?”

Applications of AI

Image credit: Alashi / Getty Images (Image has been modified)

Looking at the hundreds of startups at YC Demo Days, I wonder if I actually recognize the patterns, or if my brain invents them in a kind of pareidolia of business plans, like a battle between monotony and coffee. I’m not always sure if there is. But this year’s theme was very clear. perhaps. “

Indeed, today’s AI models are better than yesterday’s or last year’s. However, time and time again we have seen these systems demonstrate well and then fail under systematic requirements or as tools that deliver reliable and reproducible results.

It’s hard not to see this batch as a prelude to a wave of AI-powered excavatorware. Pick a use case, tweak the available models a bit (no one really builds their own), pick out some good examples with screenshots, and add a prefab UI. Congratulations, you have become the first AI social media content generation platform for independent bars and restaurants in the Middle East and North Africa. Buy hundreds of 5-star reviews and get started right away.

It’s not that restaurants in Cairo and Beirut haven’t now had handy tools to gain traction online and attract new customers. Having the currently existing AI do something is like acknowledging that it doesn’t matter.

Creating an AI-powered conversational agent to answer phone calls in your business is well-framed as a way to not lose customers. But what do customers think when a company calls them and decides they should embrace AI? Personally, I hang up and try others. What about trade workers who get AI calls to make appointments? Same thing.

Noticing that your emails are so easily “personalized” by AI is like being told “you don’t have to personalize your emails”, but we want you to think so. Do you feel cheated? It’s a systematic fraud against customers.

If your first interview with the company is with a conversation agent or someone who is clearly reading a queue generated from a knowledge base or something like that, then who is on the team and what is the size of the part for the installation? Do you feel like you’re adjusting? You don’t even deserve the full attention of a competent human being.

It wasn’t necessarily the vibe we got from all the AI ​​startups in this YC batch, but it certainly was the vibe we got from some AI startups. Below is a list of some (!) companies I wrote down that said “AI could do it”.

  • type – AI-first document editor.
  • Iliad – Generate game art assets.
  • layup – Build app-wide workflows with one-line commands, such as new employee onboarding.
  • nucleus – AI-powered onboarding orchestration that understands the “essence of the business”.
  • hadrius – SEC compliant robo-advisor.
  • speedy brand – Generated marketing content for SMBs.
  • Kwazel ・Language learning with an AI tutor.
  • Booth.ai – A generative AI “photographer” for e-commerce.
  • Skack – Natural language accounting tools.
  • Veriai – Create ChatGPT app as a service.
  • semantic – Financial news insights “enhanced” by AI.
  • Credal.ai – ChatGPT-like interface for employees to reference company documents but protect trade secrets
  • Anti-fog – Add AI data assistant to your app.
  • link grep – Suggest things from our knowledge base and add them live in your browser to chats and notes.
  • sail – vending emails.
  • iflow – Automated market research based on reviews and feedback
  • Tenner – Convert your knowledge base to a custom LLM.
  • true wind – AI-powered bookkeeping and financial processes.
  • flare lab – Collect insights from customer service call data and emails.
  • Just Paid – Automate bill payments and catch overpayments to vendors.
  • Khyber – Automate insurance industry tasks such as answering questions and underwriting.
  • Mel – A platform for training your own LLM.
  • same day – AI that calls and makes appointments with workers such as plumbers and roofers
  • Zenfetch – Live analysis of customer calls to uncover topics.
  • synchronously – AI that analyzes customer emails.
  • Pair AI – Video courses generated using AI.
  • hiding – Automation of electronic health records.
  • Avoca – AI receptionist answering missed calls over SMB.

Until about 30 seconds ago, I was actually adding my thoughts about the company to these brief and inadequate descriptions. ). Especially when many of those ideas are worked hard by the people for whom the ideas matter. It’s easy to criticize. It’s so easy that someone in the summer batch might try to automate it!

But look at that list and don’t wonder about some of the entries. or What do you really need? Doesn’t that require a lot of oversight? Doesn’t this create liability or reduce transparency? Has anyone asked customers if they want this? Who verifies and audits the results? Another AI? Who has been superseded by? Who trains people on them?

Nearly every company that presented said they went live a few weeks ago and miraculously already achieved some healthy ARR. But a few weeks is not enough to just install a major automation tool, read the documentation, and evaluate its performance and whether it’s worth the price. I can’t imagine even half of these being used, actually, by potential customers.

One example I can’t help but share: A slide from a generative marketing image company showed the following prompts for the system to work. Our Classic Ketchup is made exclusively from sweet, juicy, red ripe tomatoes and is the signature rich, rich flavor of America’s favorite ketchup. AI Copy: SWEET & JUICY KETCHUP FOR ALL! If I was a Heinz marketer and it was in a given demo, I would stand up, thank you for your time, and open the door.

Some companies have admitted to turning around in the middle of their program and just recently wrote the first lines of code for this new application. Of course, the adventurous and freewheeling nature of early-stage startups must be taken into account. That’s part of the fun and excitement of this space. But do these companies really feel “innovative”? (“Cute…here, try fintech.”)

I know you’re underestimating the amount of work it takes to build the most perfunctory AI-powered B2B SaaS service, but many of these rely on someone making an API available and everyone doing it. Feels like an old hackathon trying to squeeze in. Change to the most realistic application in hopes of getting a $1,000 gift card from something like SAP. There is joy in the process of making, but the result is not just that.

Perhaps I’ll be proven wrong when one of these companies becomes a unicorn and everyone laughs at TechCrunch’s writers for suspecting them. But I can’t shake off the worry I felt when I heard one founder after another. I’m confident that their AI can do better.

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *