Helping founders earn with AI https://xitter.cf/t.co/qOchdz8Fiu (yc f24)
NYC
Joined December 2011
- Tweets665
- Following881
- Followers536
- Likes1.5K
This is one of the biggest lessons I learned from working with @tomerlondon for years at Gusto.
Building products is an act of love and service. If you don't respect your customers, how long can you really last?
Don't listen to Dylan.
Respect your customers. Admire them. Even fall in love with them.
Building a great product requires insane attention to detail. But how do you know what details matter to your customers? You have to CARE enough about them to go through the immense pain of trial and error.
You have to care so much that it was all worth it even if you didn't figure it out.
Today is the deadline to apply to @ycombinator.
If you are on the fence - just do it. Nothing to lose, but so much to gain
the best advice for founders used to be "just charge for value; keep it simple."
in ai era just charge for value means adding markup to your inferencing, because under the hood you are upselling the raw intelligence
We must enable millions of people to earn money with AI.
It all hinges on how easy it is to build, monetize, and distribute agents.
Happy 250th birthday America.
I came here 15 years ago with a dream, and still get to live it every day since.
America is the only country build on idea, that anyone, regardless who they are, can become anything they want.
Feeling grateful to be a part of the great experiment.🇺🇸
Once subsidies end, 40k wouldn't feel like a crazy investment for a long term savings
Wow, GLM-5.2 at home at 80tps for "only" $40,000. github.com/jamesob/local-llm
Obviously still out-of-reach for most, but getting close to being cost-effective for the tokenmaxxers. A year ago there would have been another digit on this number. Where will we be next year?
irony aside, there might be 2 very important implications
1. some md files bring in millions of arr
2. there are so many more people on earth who can write very valuable md files, if we give them tools to productize it into agents
Prediction for @ycombinator S26/F26 batches:
Two largest categories of startups:
- agents
- agent infra
adding oss models for vertical ai companies is crucial for a few reasons:
- lower blended COGS, higher margins
- can allow for a mode generous free tier/higher CAC
- data on how different model perform on your workload
- reducing long term dependency on frontier models
earning money with ai is easy
build your agent
add tools
configure skills
set up 12 mcp servers
set up model routing
add support for oss models
ai sdk is wrapped 3 times for observability, billing and memory
why are all my thinking traces are in chinese
drop out of stanford
get into yc
tweet "dm for early access"
post on hn
all hn comments are "why does this exist, its so easy to do this with open claw running on my bare metal herzner box in bulgaria with eu data residency"
add credit-based billing
rewrite everything in rust
pivot to enterprise
make $175.50 MRR
this is what made Midjourney successful - people learning from each other different ways to prompt.
same applies working with agents, teammates can learn a lot from each other
if you are app layer company, you should switch to glm and run your evals asap. so many people are shocked with the performance
over the weekend, i swapped all of my loops to glm from opus & there isn't a noticeable difference in quality
the era of cost optimization is upon us. efficiency is now the bottleneck rather than intelligence
the machine gods are capable of more than we are asking of them. we need to limit test them more & expect the impossible when we prompt
TOKENMAXX
McDonalds CEO - consuming undercooked whole foods is a "very dangerous path" for most of Americans
BREAKING: Anthropic CEO says open-source AI is moving down a “very dangerous path.”
Readers added context they thought people might want to know
Anthropic CEO Dario Amodei made this statement during a US Senate hearing in July 2023. blumenthal.senate.gov/newsroom/press… techpolicy.press/transcript-sen…
every ai app layer company will go through 4 levels:
1. pre pmf - use frontier models to prove utility and value
2. post pmf - use a mix of models to drive down cost
3. scaling - start fine tuning oss models on your logs
4. train your own foundational model. Only @cursor_ai and @Base44 are on this level
Base 1 is live – Base44's own model, trained on real builder data, so it already knows what you're trying to build.
Today, Base44 becomes the first app-creation platform with its own AI model, and it's already live in production, serving builders right now.
We built it and trained it on tens of millions of real interactions on the platform. That means Base 1 learned exactly the kind of work builders do on Base44 every day, so it understands the shape of a real app from the first prompt.
For builders, everything you already use stays exactly where it is. Base 1 joins the lineup, and the platform automatically picks the best model for whatever you're building. You describe what you want, and the right model goes to work behind the scenes.
Because it's trained on real Base44 builds, Base 1 is built for the full arc of making an app, from a plain-language idea through the working backend that runs it. It already holds its own against the leading models, and it gets better every time someone builds on Base44.
Owning our own model means we can keep improving it for the exact things builders do here, and shape where it's headed next. This is the first of many steps, with bigger models and faster iteration on the way.
prediction: ai agent economics will settle on an interchange model splitting revenue:
50%-80% - to inference providers
10-20% - to harnesses, agent makers
5-10% - to external tools, plugins, skills
main question - who will be Visa and Mastercard of this era?
crazy to see our yc batchmates, and some of the earliest @afternoonHQ customers, joining forces.
@victorxheng @jahooma pushing the notion what it means for ai to be accessible
