Jayden Kim

Posting daily of open-source finds, practical workflows, and notes from my own projects.

[Jayden Kim @happybiztech]

Ever used an app and thought, “How did they build that?”

REA gives your AI agent tools to investigate software, even without its source code.

Pick one feature. Follow the evidence.

I wrote up where I’d start. Article below.

[Jayden Kim @happybiztech]

Ever used an app and wondered how one feature actually works?

I found an open-source project that gives AI agents tools to investigate software. Here’s what REA can help uncover, its limits, and why it caught my attention. https://t.co/RyUaXGehvc

[Jayden Kim @happybiztech]

@its_maddy_a @ZeroGPU_AI how are you deciding when the small model should hand off? I'd want to compare the cost of finishing a task after retries and fallbacks. curious what you trust more here: the model’s confidence or a separate check.

[Jayden Kim @happybiztech]

@zcl_2014 I’d publish task success alongside replay coverage, with missing recordings separated from agent errors. Otherwise, a model could look worse simply for exploring more. How do you compare agents that take different valid routes?

[Jayden Kim @happybiztech]

@AfterQuery would test giving the agent the same urgent work queue as smart-triage. does the gap shrink? that could help separate failures to notice what needs doing from failures to choose the right action.

[Jayden Kim @happybiztech]

You can ship a product in a weekend and still have nobody to sell it to.

That’s why Explee caught my attention.

Give it your website. The company says its agents research your market, find potential B2B buyers, write personalized outreach, follow up, and send interested prospects your booking link.

The interesting part is the work between “here’s a list of leads” and an actual sales conversation.

For a solo founder, I’d want to know:

Can it find ten companies that fit what I’m selling, and explain why each belongs on the list?

That’s the first test I’d run before letting it send anything.

A personalized email to the wrong company is still the wrong email.

If finding customers is the part of building you keep postponing, this is worth a look.

I’d judge it by prospect fit and useful conversations.

Which would you delegate first: finding the buyers or writing the first email?

https://t.co/IOy7dGV79A

[Jayden Kim @happybiztech]

@shiri_shh Google also a significant player in the space

[Jayden Kim @happybiztech]

@ConsciousRide would use new jane street agent skills repo to start. wrote about it on my page

[Jayden Kim @happybiztech]

@kunchenguid interesting take. Cal Newport had a similar prediction of where the world was headed 10 years ago in his book Deep Work. Said people who are going to survive in the new economy are those who can learn hard things quickly and apply them.

[Jayden Kim @happybiztech]

@Polymarket AI seems to be headed in the direction to become the place you buy things. If you choose and book a trip inside ChatGPT, a travel site might still process the booking but lose the customer relationship. Wondering who gets to decide which options the AI shows you.

[Jayden Kim @happybiztech]

@suraj_sharma14 If you’re starting with RAG, write 20 questions about your docs, including a few they can’t answer. Then look at what the app retrieves when it gets one wrong. Did it pull the wrong passage, or get the right passage and still mess up? Those need different fixes.

[Jayden Kim @happybiztech]

This is f*cking worth trying.

Jane Street has a public AI skill that makes you prove you understand something before moving on.

Think about your last AI conversation.

The explanation made sense.

Could you explain it back tomorrow with the chat closed?

That’s the question behind its open-source “socratic-tutor” skill.

Here’s how it works:

step 1 → Check your starting point — find out what you actually know

step 2 → Ask one question at a time — make you think through the answer

step 3 → Probe weak answers — find the hole in your reasoning and keep digging

step 4 → Revisit earlier material — check whether you can recall it without being fed the answer first

step 5 → Finish with a mixed assessment — test whether you can put the pieces together

You can load it into a compatible agent or paste the skill instructions into a chatbot.

Then pick something you’ve been meaning to understand.

Statistics. Investing. A concept you keep looking up at work.

And try answering before asking for another explanation.

The uncomfortable moment when you get stuck?

That’s the part worth finding.

I wrote a breakdown of the skill, how to try it, and what to watch for.

Bookmark this for your next study session.

Full article below ↓

[Jayden Kim @happybiztech]

This Jane Street Repo Completely Changed the Way I Learn with AI https://t.co/s4tzWPjDc7

[Jayden Kim @happybiztech]

This Jane Street Repo Completely Changed the Way I Learn with AI https://t.co/s4tzWPjDc7

[Francis Davidson @FDavidsonT]

https://t.co/F3Wq6IJ6rJ

[Jayden Kim @happybiztech]

@karanwalia_ai @MIT_CSAIL @MSFTResearch @Cambridge_Uni @CarnegieMellon COLD

[Jayden Kim @happybiztech]

If your engineer is telling you to build another rag-based AI tool, you should probably fire them immediately.

These data-centric architectures are only a one-way street for insights & visualizations.

This might not seem that bad, but in-reality, if you’re here, you more than likely, are falling behind in frontier AI application for your business.

When building AI-driven automations, the focus should be centered on your processes rather than just the data.

When looking into the applications of AI for your company, it’s very easy to get sucked into being heavily data-centric.
However, this only helps aggregate information, creating dashboards that offer occasional insights. When you build in AI like this, you system becomes just another tool, not something that is designed to power any critical front-line activity.