AI will change the way we shop - from where we find products to how we evaluate them, when we buy, and much more.
What types of purchases will be disrupted, and where does opportunity exist in the age of AI?
More from me + @arampell 👇
To start - what are the categories of commerce? (for consumers)
We divide them by level of consideration, from impulse buys to life purchases.
These have vastly different processes - you don't buy a new backpack and a car in the same way - which means the way AI touches each purchase category will vary.
Some thoughts on how this might shake out:
1) Impulse buys - the candy bars you pick up at the checkout counter (or their digital equivalent).
You don't do a lot of research in advance, so it's tough for AI to play a role in your shopping process.
But the algorithms on social apps will continue to improve + target you with more relevant impulse purchases (like that cat-shaped water bottle or $15 t-shirt from your favorite show).
2) Routine essentials - things you buy regularly (groceries, household supplies, pet food).
You have products you know and love. But AI can help find where you can get the best price - and maybe even purchase on your behalf if it spots an incredible deal.
Products like @camelcamelcamel, which alerts you to price drops on Amazon items, are early examples of this.
3) Lifestyle purchases - things you don't buy every week, like a wedding guest dress or a nice briefcase.
You're going to want to research and consider a few options. What if an AI agent does the grunt work for you and come back with a summary of what it recommends and why?
Products like @perplexity_ai Shopping are making progress here.
4) Functional purchases - things you use regularly that serve a practical purpose in your life.
Think a couch in your living room, your laptop, or a bike you use for commuting. These things need to hold up!
In addition to having an AI agent do research, you'll also probably want to talk to an expert about your unique needs or debate the pros and cons of various options.
This "expert" could also be AI...imagine an LLM trained on conversations for a specific product category, with integrations to access the most up-to-date data.
5) Life purchases - think buying a house or a car. Or picking a college.
These are highly considered purchases that happen only a few times in your life. And it's unlikely you'll fully outsource these decisions to AI.
However, it's very possible you'll use an AI coach to guide you through the process - from structuring your decision-making process to debating various options.
@a16z @arampell Thanks for reading!
If you're building in this space, we'd love to hear from you - you can DM us here (@venturetwins, @arampell) or email me (jmoore@a16z.com).
Fun experiment: you can now automate content production with Claude Cowork.
I gave Claude a browser and asked it to: 1) Find a NeurIPS best paper 2) Write a thread on it 3) Generate graphics on Krea 4) Validate its work with ChatGPT.
Shockingly, it worked! How to do it 👇
You start in the "Cowork" tab in the desktop app.
Claude will open a browser and search, so you can give a vague or very specific prompt (like I did) and it will find context.
I mentioned I wanted image gen on Krea because I got sick of watching Claude search random sites 😂
In terms of the actual written content - answer Claude's Qs about who your audience is + how long it should be.
This will heavily influence results!
It will then browse the relevant sources and open up an internal scratchpad to save down insights and start drafting the thread.
My favorite paper this year: "Video models are zero-shot learners and reasoners"
It illustrates that video models show emergent visual reasoning at scale - they can solve vision tasks they weren't trained for.
This may be the "GPT moment" for vision. Let's break it down 👇
To start - why believe that video models might develop visual reasoning?
A similar thing happened in text. We used to train specific models for each task - but now, LLMs have general language understanding and can tackle lots of tasks that they weren't explicitly trained for.
It's feasible that video models may do the same at scale.
This paper measured 18k+ videos generated by Veo 3 across both qualitative and quantitative tasks.
It found that Veo can perceive, modify, and manipulate the visual world (starting from image + text prompts) - showcasing early reasoning skills that it wasn't explicitly trained for.
Three years ago, I went down the rabbit hole on AI x creative tools - and it's been incredible to see how quickly this space has evolved.
Thrilled to bring together the best founders + creators to share learnings + try new tools!
Our curated group of creators will spend the day in demo sessions featuring the best tools across modalities: @elevenlabsio, @theworldlabs, @krea_ai, @hedra_labs, @ideogram_ai, @heyglif, @LumaLabsAI, @ViggleAI
We're very lucky to have these founders + product leads join us.
@elevenlabsio @theworldlabs @krea_ai @hedra_labs @ideogram_ai @heyglif @LumaLabsAI @ViggleAI Exciting launch today - @LumaLabsAI's Ray 3 dropped, and the team is here to give our creators a demo + chat through the model strengths.
It's a reasoning video model that can generate studio-grade HDR 🤯