Richard Li

Does anyone know how to use Codex's agent orchestrator, e.g., github.com/openai/codex...?

#AI

@simonwillison.net have you seen this?

Don't Think Twice (It's Alright) - Grant Berness Cover of Bob Dylan YouTube video by Grant Berness

Proud uncle moment.

My 16-year-old nephew, Grant Berness, is an aspiring musician and recently recorded a Bob Dylan cover. He set a simple goal for himself: reach 1k views.

If you have a few minutes, give it a watch and share: youtu.be/QHAtgRs7r0s

This is the agentic runtime stack—what’s needed to run autonomous, goal-directed agents in production.
#AgenticAI #LLM #AIinfrastructure

The Agentic AI Runtime Stack | Wing Venture Capital There’s no single architectural stack for agents, as they have multiple complex components. We see three distinct stacks for agentic AI: a model training stack, an inference stack, and an agentic runtime stack.

#Agents aren’t the future—they’re already here.
But building them? That takes a whole new stack.

🧠 AI
⚙️ Durable execution
🧱 Frameworks
🗂 Context
🛠 Actuators

Check out the breakdown (with @jflomenb and @Wing_VC):
🔗 www.wing.vc/content/the...

The Agentic AI Runtime Stack | Wing Venture Capital There’s no single architectural stack for agents, as they have multiple complex components. We see three distinct stacks for agentic AI: a model training stack, an inference stack, and an agentic runtime stack.

Context Is King 👑. Smarter agents need richer context—not just prompts. They aggregate context from multiple sources:

🧠 Knowledge (vector databases)
💾 Memory (short/long-term)
🌐 Actuators (APIs, sensors)

Check out the full Agentic Runtime Stack:
www.wing.vc/content/the...

The Agentic AI Runtime Stack | Wing Venture Capital There’s no single architectural stack for agents, as they have multiple complex components. We see three distinct stacks for agentic AI: a model training stack, an inference stack, and an agentic runtime stack.

Most apps act fast. Agents don’t.

They pause, retry, wait hours or days. That’s why they need durable execution—resilient workflows that persist across failures.

Check out our post on the agentic runtime stack:

www.wing.vc/content/the...

The Agentic AI Runtime Stack | Wing Venture Capital There’s no single architectural stack for agents, as they have multiple complex components. We see three distinct stacks for agentic AI: a model training stack, an inference stack, and an agentic runtime stack.

Agentic frameworks are expanding fast—most started with some sort of syntactic sugar around prompts, but they've since expanded into durable execution & memory.

I teamed up with @jflomenb + @wing_vc to map the stack 👇
🔗 www.wing.vc/content/the...
#LLMops #AgenticAI

The Agentic AI Runtime Stack | Wing Venture Capital There’s no single architectural stack for agents, as they have multiple complex components. We see three distinct stacks for agentic AI: a model training stack, an inference stack, and an agentic runt...

🚀 Agents aren't the future—they're here.

From legal #AI to AI software engineers, #AgenticAI systems are reshaping how work gets done. Building them requires a new stack: the Agentic Runtime Stack.

🧠 AI
⚙️ Durable Execution
🧱 Agentic Frameworks
🗂 Context
🛠 Actuators

www.wing.vc/content/the-...

If you're working in this area, would love to chat!

I also see how building a domain-specific reasoning model that outperforms the generalist models in your domain is entirely possible!

What I think is most interesting is that building these models is now within the reach of engineering/practitioner teams, and not just a research-type team.

Engineering reasoning LLMs: Notes and Observations Engineering reasoning LLMs: Notes and Observations

Reasoning LLMs like OpenAI o1 and DeepSeek-R1 have shown how #AI can tackle complex problems by breaking them down. I've been doing work in this area, and have written up my notes here: www.thelis.org/blog/reasoni...

The era of #agenticAI is here.

#AIagents are replacing manual workflows—making decisions, taking action, and scaling fast.

Learn how to design and implement the next generation of AI-powered services.

Register now: register.gotowebinar.com/register/907...

#Java #AI @infoq.com @rdli.bsky.social

The missing piece of software product requirements: Price product requirements should always have a price

Setting a price that you're willing to pay is a critical component of software product requirements -- but many PMs don't communicate the price. This seemingly simple bit of information enables the entire org to make better tradeoffs during the software process.

www.thelis.org/blog/set-a-p...

TL, DR: If you're joining today, ask for RSUs. If you joined a few years ago, congratulations!

The #AI market is 🔥. Anthropic is raising another $2B at a $60B valuation when their run-rate revenue at the end of 2024 was $1B. I ran some back-of-the-envelope math to figure out under what scenarios Anthropic stock options are valuable.

www.thelis.org/blog/anthrop...

Fitting Facts to the Narrative at The Washington Post Link to: https://www.washingtonpost.com/technology/2024/05/02/apple-sales-ai-iphone-earnings-regulations/

@daringfireball.bsky.social on less than a year ago, you said "While I’m being grumpy, I’ll even take issue with the notion  ... that Apple is “behind in AI”? daringfireball.net/linked/2024/.... Do you still agree?

AI Developer Frameworks and the evolving AI infrastructure ecosystem AI Developer Frameworks and the evolving AI infrastructure ecosystem

Thanks! I realized i didnt do a freat job explaining myself on franeworks, so i wrote a more cogent (hopefully) take here: www.thelis.org/blog/ai-dev-....

A common response I got to my post was "Try $OTHER_AI_FRAMEWORK." I wasn't trying to critique @langchain.bsky.social or @llamaindex.bsky.social. AI dev frameworks in general are solving the wrong problem, & the real challenges with #ai apps are elsewhere:

www.thelis.org/blog/ai-dev-...

I wrote about this because many of the problems I encountered were not what I would characterize as a small scale problem.

7 Lessons from building an AI application 7 Lessons from building an AI application for a year

I've been building an #ai application for a little while now, and wrote up my 7 macro takeaways about building an AI app that I didn't know when I started.

www.thelis.org/blog/lessons...

Rain Saturday afternoon, break Sunday morning. Snow/mix arrives Sunday afternoon. Storm ends Monday morning. Still 48 hours or so until snow underway. Expect the bands to shift as we get closer to storm. Cold air comes in before so fluffy snow can pile up faster. #Snow

The vendor email that wishes me “Happy Holidays” or “Happy New Year” is … just irritating. Don’t send it.

THIS IS HUGE! Researchers at TAU and the University of Lisbon have developed a groundbreaking nano-vaccine against COVID-19, delivered via nasal spray WITHOUT requiring cold storage. Preclincial studies show the vaccine was effective against ALL major SARS-CoV-2 variants. Let’s talk about that! 🧪🧵⬇️

This is 💯

I woke up one morning on blue sky and discovered a swarm of cloud native refugees.

I attended a meetup in Seattle and got to spend time with @bnewbold.net and @pfrazee.com from the Bluesky engineering team.

I walked away with an understanding of the role DIDs and PDSs play in the grand scheme of things, and why the most exciting work sits higher up the stack. I almost unretired.

Ranking worst to best hyperscaler cloud IAM.

3. Azure (worst)
2. Amazon (less worse)
1. Google (least worse)

I know RBAC is complicated and there’s legacy but why can’t the documentation be straightforward at least?

I'm watching some folks reverse engineer the xz backdoor, sharing some *preliminary* analysis with permission.

The hooked RSA_public_decrypt verifies a signature on the server's host key by a fixed Ed448 key, and then passes a payload to system().

It's RCE, not auth bypass, and gated/unreplayable.

I’ve been thinking a lot about how to analyze data (dashboards can’t be the right answer!) and wrote up some thoughts here: www.amorphousdata.com/blog/data-is.... #data