Building @coniferbuild (YC S26) | Tech × finance @ Princeton.
Joined January 2026
- Tweets130
- Following28
- Followers82
- Likes577
Now that the blue wave is taking over midterms, it looks like the red wave is taking over the stock market 😭.
Many people think that we're in a bubble, and that once these congress people are assuming office they will put a ban on Artificial Superintelligence, popping the bubble. With how dangerous AI could become, this might be the smart move; however, we still need to consider the consequences of halting AI development.
China open weights models are only a few months behind the current frontier labs, and if we were to halt development that would enable China to become the frontier.
Because of this, I personally don't think the bubble will be popping just yet as long as China is still considered a threat in the AI space. On top of that, Anthropic's IPO is the most anticipated IPO in history, so imo, it is unlikely the bubble would be popping before then. As a result, I am not selling my stocks just yet; I believe AI as a whole is still undervalued.
The only way that I can see AI development halting is if we have an international treaty with China and all of the other countries to have a united effort of slowing down AI development. However, I'm not sure how likely that is?
Lowkey a smart move by YC. In 10 years when every single software companies can be one-shottable by Fable 100, the only industries that still might be able to survive are those targeting your dopamines (casinos, gaming, etc).
Pay for Conifer with an ernd card and get 5% back on every charge!
@ernd_ai is now partnered with Conifer (YC26). Cardholders receive 5% back on all spend.
Conifer is a free gateway and router for AI models, built for enterprise needs. Change a single API key to access over 240 models, with each request routed to deliver maximum intelligence at the lowest cost. Built-in fallbacks and failover ensure provider outages never reach your users, and teams can create custom routing pools that mix in their own fine-tuned or self-hosted models
Request a card for your business at ernd.ai
As AI makes things easier to build, I believe creativity and entertainment will start to matter more and more.
People will still care about how a product feels rather than just how good of a product it is, especially if they enjoy spending their time on a product. Turning AI development into a Club Penguin game is such a great idea. I wished I thought of this. It for sure combined my childhood nostalgia with the responsibilities of developing in the current AI ecosystem. Hopefully more people will think of similar things and we can make coding environments fun.
This week and upcoming week will be one of the craziest times in AI. Not only are rumors of Astra 6.1/Bel 6.2 and Fable 5.5/Opus 6 coming out soon, but the benchmarks on internal models leaks seems to be very promising.
What surprised me the most recently is the development of Gemini 4 models leading the frontier. Carbon 4 seems to be better than Opus 5.5, and leaks of Gemini 4 Xenon says that it is expected to outclass Fable 5.5 (a model that hasn't even been released yet?). OpenAI might also might not be behind Anthropic either since another leak pointed out how their Gpt 6.2 Bel model was rumored to score a 72 on AA. (To put that into perspective, Opus 5.5 to GPT 6.2 Bel is a 14 point jump, that is a higher jump than Kimi K3.1 Pro and Opus 5.5! (12 points)).
As Anthropic is preparing for their IPO, I doubt they'll let the another labs take control of the frontier though...I'm excited to see how Fable 5.5 and Opus 6 will compare to Gpt 6.1/6.2. Who knows, Qwen 4 is also releasing soon, and maybe it'll surprise us with its benchmarks!
Loc retweeted
Conifer is partnering with ernd, the credit card for the AI stack.
Pay for Conifer with an ernd card and get 5% back on every charge.
I’m super excited to be a part of this project and share Parallax with everyone.
I usually have a ton of agent sessions running at once, and for the longest time I was the one keeping track of all of them. I was also constantly juggling subscriptions, and so whenever one would hit its limit mid-task, I would have to stop, switch accounts and pick it back up. On top of that, every time a new model dropped I’d move my whole workflow (Codex when GPT-6 Astra came out, then back to Claude when Opus 5.5 came out).
The separate apps, separate sessions, separate everything were a huge hassle. Parallax fixed all of that for me. When one subscription runs out, it moves the work to the next one and my agents just keep going. And every model is in one place, so switching to whatever just came out takes a second. But the thing I’m most excited to share is Captain. It keeps track of all my sessions for me and makes sure each one stays aligned with what I’m actually trying to do. If an agent starts drifting, Captain steers it back so it answers the prompt and hits the objective. I finally don’t have to watch every session myself.
We just launched, and I’d love for you to try it
We''re excited to announce we are launching Parallax!
We think one of the biggest challenges facing coding agents today is agent-developer alignment. There’s an undeniable asymmetry between a one-line prompt and hours of agentic work.
A core problem is that an agent’s next action depends on the context it has, but agents discover context as they go. That means they make some of their biggest decisions early, when they know the least.
Our solution is to review the agent’s work while it’s happening, not after. Parallax runs a dedicated reviewer alongside each agent, watching its trajectory as it works. When the reviewer spots a bug or a wrong turn, it steers the agent mid-task. It also shares what it learns with your other running agents, so every session gets smarter and the overall bug rate goes down.
In our testing, this caught more bugs than traditional code reviewers, and improved our internal alignment scores by 50%.
The best part is this works with your existing harness and subscriptions! We’ve also added a bunch of quality-of-life features that our pilot teams and we ourselves have found useful:
• Immediate access to every model through Confier’s gateway
• Automatic switching between your ChatGPT, Claude, and other subscriptions
• Subscription use on cloud machines
• And my personal favorite: GrokBot, Hermes, or your other persistent agents can natively coordinate and orchestrate all of your sessions
We’ve left a few more features for you to discover, so tell us what you find!
conifer.build/parallax/
Loc retweeted
We're officially launched on Product Hunt!
producthunt.com/products/par…
Loc retweeted
We''re excited to announce we are launching Parallax!
We think one of the biggest challenges facing coding agents today is agent-developer alignment. There’s an undeniable asymmetry between a one-line prompt and hours of agentic work.
A core problem is that an agent’s next action depends on the context it has, but agents discover context as they go. That means they make some of their biggest decisions early, when they know the least.
Our solution is to review the agent’s work while it’s happening, not after. Parallax runs a dedicated reviewer alongside each agent, watching its trajectory as it works. When the reviewer spots a bug or a wrong turn, it steers the agent mid-task. It also shares what it learns with your other running agents, so every session gets smarter and the overall bug rate goes down.
In our testing, this caught more bugs than traditional code reviewers, and improved our internal alignment scores by 50%.
The best part is this works with your existing harness and subscriptions! We’ve also added a bunch of quality-of-life features that our pilot teams and we ourselves have found useful:
• Immediate access to every model through Confier’s gateway
• Automatic switching between your ChatGPT, Claude, and other subscriptions
• Subscription use on cloud machines
• And my personal favorite: GrokBot, Hermes, or your other persistent agents can natively coordinate and orchestrate all of your sessions
We’ve left a few more features for you to discover, so tell us what you find!
conifer.build/parallax/
Loc retweeted
Gemini 4 Argon has been blowing us away with its performance. If you want to try it for yourself, it is available now in Conifer's gateway!
Loc retweeted
Where our "cut token costs by up to 80%" figure comes from. Here are four simple things we do on every call.
1. Most queries are easy. Classification, extraction and document OCR now cost 1/10 of what they did 6 months ago.
2. The hard calls stay on the strongest model. Set your own quality floors and ceilings, depending on whether you'd rather minimize cost or maximize intelligence.
3. Prompt caching is managed for you. Learned behavior allows for more frequent model switches without losing your cache.
4. A dashboard built for observability. Create projects, mint keys under each one, and track exactly where your spend goes.
None of this needs a new SDK. Mint an API key and start today.
I can not believe they are forming an AI task force. As a founder, I’m excited that AI lets small teams build bigger companies. However, I worry about what happens when that growth doesn’t come with more hiring. A lot of my friends are still in college, graduating into one of the worst job markets of our times.
My hope for the AI task force: make powerful AI affordable enough that more people have a shot at starting a company. We see new models push upon the pareto frontier as these models gets cheaper and more intellegent. Using these models to start your own company, in this job market, might be the safer move than trying to get hired.
“I am announcing the formation of the Super Intelligence Force (SIF). The Super Intelligence Force is tasked with coordinating the effort of the Federal Government to ensure that America continues to lead the World in Super Intelligence, which many say is bigger than the Industrial Revolution, and the Internet, and will protect the interests, and improve the lives, of all Americans.” - President DONALD J. TRUMP 🇺🇸
