Rachel Decker

[Jamie Voynow @voynow]

We’re hiring a builder/operator hybrid to work on agentic quant research & trading systems

The Work:
- long-running multi-agent systems
- live trading workflows + terabyte-scale data
- high impact, broad scope

The Team:
- small, very strong, self-funded
- traders + engineers running live systems
- years of proven edge
- fully remote

don’t send a resume
dm me with links to what you’ve built

[Jamie Voynow @voynow]

Betting on LLMs betting on markets

Call that a derivative of a derivative

[Jamie Voynow @voynow]

I am officially a professional poster damn that’s crazy

[Jamie Voynow @voynow]

cushy $130k swe job in nyc

[Jamie Voynow @voynow]

I’m letting gpt5 drive prompt eng and am genuinely so so shocked at how good the results are. Kind of insane. I’m starting to think that the DSPy philosophy was right all along

[Greg Brockman @gdb]

have gpt-5 write the prompts for you:

[Jamie Voynow @voynow]

A lot of new followers over the past month or so. Figured it’s time to reintroduce myself:

Hey! I’m Jamie.

Here’s how I got here
> Started studying ML when I was 19
> Broke into Data Science after undergrad
> Pivoted from researcher to engineer (Being a builder is way cooler imo)
> Was early to AI eng, hacking in my free time
> Landed on an AI eng team at BigCo
> Was grinding to break out of the wagie cage
> Launched a ton of projects with little engagement
> Then one afternoon I built https://t.co/14tG6uY7PG and my life changed forever
> Was the catalyst for massive follower growth
> And opened doors to be at the bleeding edge of agent eng @ House Kling

This journey has been insane and I feel so lucky & grateful tbh

If you are early on a path similar to mine: being on X is a good start, but I cannot stress enough how important it is to *post your work*. Being able to point to it and say “Yes I can build *this* for your company” will get you paid

[Jamie Voynow @voynow]

seriously though what is stopping you from coding like this

[Jamie Voynow @voynow]

Another day of being a pretentious coffee shop dbag. I’m kinda starting to like this lifestyle

[Jamie Voynow @voynow]

Agent Design Pattern: Parallel Rollouts

Inspired by Tree-of-Thought [1] and @corbtt's Universal Reward Function [2], lately I've been using a best-of-n pattern (dubbed “parallel rollouts” internally) and seeing consistently strong results

When designing an agent, retry-on-failure is generally a good practice, but it’s costly (in terms of latency). Instead, let’s just assume the agent will act sub-optimally:
> Let the agent perform N rollouts concurrently
> Grade each rollout via LLM-as-a-judge
> Select the best one
> No retries needed

It’s a simple tradeoff: higher cost for lower latency - worth it for high margin agentic tasks (not for everyone)

Note: This idea of search and selection is not novel (see MCTS, AlphaGo, etc) but in the context of agents branching is underused

[Jamie Voynow @voynow]

Life Update!

- Turned down an ML eng offer @ Meta (no it wasn’t $100m)
- Stumbled upon the opportunity of a lifetime building AI Agents for Quant
- Quit my job (was a great job, sad to go)

Now it’s time to go deep on Agents/Quant. Who are the best follows in this space? Lmk

[Jamie Voynow @voynow]

Annnnnnd if you would like to see this project continue: every dollar donated here will go directly into my OpenAI account for more tests: https://t.co/GwaoqYAVZF

[Jamie Voynow @voynow]

Poker Bench results are in.

I simulated 100 games of Texas Hold 'em between:
- 4.1-mini, 4.1-nano, 4o-mini (base)
- 4.1-mini, 4.1-nano, 4o-mini (with reasoning)

27 million tokens later, here’s what I learned:

[Jamie Voynow @voynow]

Confession: I regret vibe coding so much over the past year or so and am actively trying to depend less on LLMs for writing code. I can't help but feel like my focus, debugging skills, and syntax muscle memory is worse off. Trading low level abilities for speed and product sense is good in the short term, but is not sustainable over a long career in technology.

[Jamie Voynow @voynow]

voynow(dot)ai is live. Investors? don't all come running at once

[Jamie Voynow @voynow]

How to Vibe Code without getting lost in the sauce

My take on the stages of Vibe Coding as software products evolve from demo to production

Demo: Pure Vibes
Get the demo out as fast as possible despite the grossness of the code. Validate quickly. A landing page, a notebook, or a simple API will do. You’re accruing tech debt at a rapid pace, so finish this stage fast (faster than an MVP).

Iterating: Vibes Plus Taste
Time to iterate and add features. Patterns will emerge, lean in, refine, and enforce structure. You’re still allowed to use AI-generated code, but you should resist the urge to totally vibe out, the patterns built today will make or break your velocity tomorrow. Practical advice: double down on extendibility with strong typing and clean interfaces

Production: 10x Vibe Engineer
Now that patterns are stable, codegen becomes EVEN MORE reliable. Write your code like a 10x wizard dev, and your AI will output 10x wizard code (LLMs are very impressionable). Queries like “Add feature Y accessing data from table X” are incredibly powerful because the LLM can extrapolate from existing patterns. Frontier models are not bound by intelligence, they are context-bound, so you must ensure that you give them proper context to cook.

Start fast, iterate intentionally, and reap the rewards of long-term well-written codegen

[Jamie Voynow @voynow]

I'm a huge fan of OpenAI, but this is bad.

Evals that lock you into a single LLM provider defeat the entire purpose of an evaluation platform. In a world where LLM benchmarks and leaderboards are being gamed and increasingly meaningless, running your own evals (across models) on the tasks you care about matters more than ever.

So while I commend OpenAI for making LLMops potentially more convenient, I don't recommend locking yourself into this platform.

[Rachel Decker @rsdecker99]

@voynow Happy birthday!

[Jamie Voynow @voynow]

https://t.co/mCvPq6EXOm

[Jamie Voynow @voynow]

Annnnd if any of you sugar mommies / daddies want to fund me, I do pay for this out of my own pocket: https://t.co/GwaoqYAVZF