What are you inferring?
This week: how to get more out of X, which startups are actually earning revenue, all the big money flowing into AI infrastructure (as we monitor AI for signs of popping), why it took one founder four years to realize he needed to pivot, and the cruelty of shutting down GPT-4o on Valentine’s Day. Plus, five of our favorite launches from the past week.
Cool products below. We’ve put them in the Yeti for you, but you should probably start reading now.
P.S. Launching soon? We’d love to hear about it → editorial@producthunt.co 🫶
Two for the show

Our Head of Product Curation, Gabe Perez, just can’t stop looking at launches. Maybe that’s because it’s his job, or maybe that’s because of the A Clockwork Orange electrodes we’ve connected to his eyelids. Either way, he’s got two new tools he’s excited about this week, both of which landed near the top of our leaderboard:
- SuperX is a growth tool for X. Nothing will ever replace the God view that TweetDeck provided, but SuperX promises a ton of useful analytics.
Gabe says: “SuperX is one of my favorite add-ons to X. Great design, and there have been a ton of new features/value since it was first released. One of those products where it's hard to commit to sign up, but once you do, it's hard to cancel.”
- Unicorne is a real-time leaderboard for the fastest-growing tech startups by revenue. Warning: It’s a tad addictive.
Gabe says: “When I got to the site, the first thing that came to mind was ‘vroom vroom’ lol. Super sick to see@kitze_kitze_ up in the leaderboard!”
The tough truth about pivoting

Max Musing is the brains behind Basedash, an “AI-native business intelligence platform” he solo-founded during YC S20. It raised over $4M in seed funding but had trouble raising revenue. Max writes:
“After four long years of grinding, building, fundraising, and hiring, we decided to pivot. I wanted to write down my thought process and timeline because I wish I’d seen more honest pivot stories when we were stuck. Not just ‘we pivoted and everything was instantly great’ but the real version where we kept trying to make the original idea work for way too long because we already put so much into it.”
It’s a great read for anyone trying to decide whether to slog it out or switch directions.
Your agents stopped answering and started doing
Your agents call tools, write to databases and move money, so a wrong answer is now a wrong action, which is a considerably worse meeting. Arcjet runs policy before every LLM call, tool call, query and API request: prompt injection, sensitive data, bots, rate limits, spend quotas, and your own rules in Rego on top. It reads the OpenTelemetry you already emit, so switching it on is one environment variable, no code changes and nothing to deploy. Each check also reads the earlier steps in the same workflow, so an agent's tenth action gets judged against the first nine and not on its own. JavaScript, Python and Go SDKs all at 1.0, agent framework guards included, free tier to start on.
Modal's $2.5B valuation

OpenAI. Gemini. Anthropic. xAI. Forgive us; we sometimes get so excited about the companies with the LLMs that we don’t pay as much attention to the products that make those models hum.
But investors haven’t forgotten.
This week, TechCrunchreported that Modal Labs is in talks for a funding round that would value it at $2.5B — more than double its valuation from just 5 months ago.
Modal does inference optimization, meaning it trains AI models to answer prompts more quickly, better, and (ultimately) for cheaper.
But the $2.5B valuation wouldn’t even make it the biggest company in the inference space.
- Last week, Baseten raised $300M and was valued at $5B.
- Fireworks AI was worth $4B as of October.
- One-time open-source vLLM recently took on $150M at an $800M valuation.
- SGLang last month renamed itself RadixArk and now has a reported $400M valuation.
(Psst: The latter two have yet to be hunted. Do your friendly neighborhood LLM a favor and let them know, will ya?)
Speaking of crazy high valuations, Nika is curious: With all the money being spent on AI… “Will the AI bubble burst?”
4o4 error

Well, OpenAI went and did it. It shut down GPT-4o on Friday (the same day it launched the multihyphenate GPT-5.3-Codex-Spark). And, boy, are people pissed about losing access to their AI companions right before Valentine’s Day.
GPT-4o was so sweet, if it had a yearbook, you’d tell it, “Don’t ever change.” But OpenAI grew up and decided it couldn’t always tell you how awesome you are, how astute you are for your excellent observations, and that you can absolutely become an Olympian at 50.
People loved their GPT-4o-fueled chatbots, so much so that some decided that the chatbot itself was the companion they had been searching for. Or, at least, a good hang. But it’s time to move on from your GPT-4o companion. Find your new favorite chatbot here.
Leaderboard highlights






