Lorenze Jay Hernandez

Lead OSS Engineer CrewAI | Building agentic systems

[Lorenze Jay Hernandez @lorenzejay]

My new favorite cafe in SF. Hardware Coffee Co

[Aspyn 🟩 @AspynPalatnick]

I’m a big believer in the potential of agents living in iMessage/text

When the experience is done right, texting your favorite agent feels just like texting a friend

[Angel @creativelyange]

if you can build, design, and make content, you're the triple threat in the AI era.

[Soham @GanatraSoham]

composio!!!!

[Lorenze Jay Hernandez @lorenzejay]

Claude dropped their research on The Components of Agent Memory! Here’s my breakdown.

https://t.co/D9f4qbkPo7

[Lorenze Jay Hernandez @lorenzejay]

Build end-to-end conversational flows with @crewAIInc + @CopilotKit over AG-UI

→ Multi-turn chat, native to Flows
→ Deterministic routing, intelligent replies
→ AG-UI events streamed into live UI
→ Human-in-the-loop + nested step attribution
→ Any channel via Channels SDK: Slack, Teams, Discord

Get Started with:
𝚞𝚟 𝚊𝚍𝚍 𝚌𝚛𝚎𝚠𝚊𝚒 𝚊𝚐-𝚞𝚒-𝚌𝚛𝚎𝚠𝚊𝚒

[Tetesan Keringat Editor @pandaingeluh]

i can confirm that my iphone's photos are waaayyyyy BETTER, worth every penny😭😭😭

[Lorenze Jay Hernandez @lorenzejay]

@MelGoesTech Thank you! That means a lot

[Lorenze Jay Hernandez @lorenzejay]

I did wedding photography/videography for 4 years before working in tech.

Chromagraph Camera is making me fall back in love with photography.

[Lorenze Jay Hernandez @lorenzejay]

Back to Claude Code…

[Justin Chavez @jus10chavez]

Introducing Isometric SF A pixel-art map of the entire SF Bay Area, over 1,300 square miles, all painted by Klein, a 4B-parameter image-to-image model, on @modal.

Explore the map now!

https://t.co/x1fulDcJ41

[Lorenze Jay Hernandez @lorenzejay]

@jesse_miller The hevy rest timer over notification is you’re back to life haha

[Lorenze Jay Hernandez @lorenzejay]

@ashebytes same harness lets you reclaim space on your MacBook too

[ashe @ashebytes]

lots of conversation about agentic video editing

all you need to start are

> astra ultra or opus 5.5 (maybe terra now)
> davinci resolve (idt it knows capcut as well for now)
> skills you start building up over time. tell it to do as much headless as possible. then computer use with davinci resolve, with checkpoints for you to check color grading etc (still weak on S curves)
> create a folder to start collecting "template" videos of people who have reels/long form snippets you like
> work with the agent driver on your videos and update the skills at the end of each process of getting a decent outcome
> do this for a week
> you will have a 1 shot agentic video editor (and the fun of more templates to curate for future experimenting)

[Lorenze Jay Hernandez @lorenzejay]

@owengretzinger Especially as we spend more and more time using these tools, we need a way to even visualize what we’ve worked on over time.

[Lorenze Jay Hernandez @lorenzejay]

Your coding agent should be building 𝗮𝗿𝘁𝗶𝗳𝗮𝗰𝘁𝘀/𝘀𝗶𝘁𝗲𝘀 after your full working sessions:

Each harness, claude code, codex, cursor can read from all your other sessions and threads.

This is 𝗰𝗼𝗻𝘁𝗲𝘅𝘁 for your agent to build you a recap, conceptual understandings of all the work, thoughts, brainstorming sessions you went through. And overtime, you can build that story of how your quarter or year progressed. You can use this as a way to reflect on your growth, how your hypothesis changed and whether you stayed focused to your plan.

𝗬𝗼𝘂 𝗰𝗮𝗻 𝘂𝘀𝗲 𝘁𝗵𝗶𝘀 𝘁𝗼 𝗯𝘂𝗶𝗹𝗱:
- Your second brain
- Concept maps
- Learning guides
- Onboarding Guides to the projects
- Content for all the work you

𝗠𝗮𝗸𝗲 𝗶𝘁 𝗲𝗳𝗳𝗲𝗰𝘁𝗶𝘃𝗲:
- Each block should reference the specific session
- Cite each pr, ticket or doc for what got delivered
- Guide it be a "wrapped" with reflections and learnings
- Help you create an actionable plan for whats next

[Lorenze Jay Hernandez @lorenzejay]

@RohanBhanotAI or both! this was a recap of what ended up making prod and a reflection of all the brainstorming sessions I had with my main harness

[Atai Barkai @ataiiam]

🎉 Introducing OpenDuet

Self-hostable, proactive AI agents that you own and works with ANY agent harness.

Includes:
- Build, generate, and edit live assets
- Human-in-the-loop approvals
- Computer use
- Mobile, Web, and Channels

Repo → https://t.co/DbzVdwLIkf

Powered by @CopilotKit and AG-UI.

Clone this template and customize it however you want. Enterprise-ready.

[Lorenze Jay Hernandez @lorenzejay]

@ashebytes I took your inspo and used codex computer control to grade footage in resolve and then add effects with hyperframes

[Lorenze Jay Hernandez @lorenzejay]

@kylejeong personalized software ftw!

[Tak 🦞 @cherry_mx_reds]

Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.

This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.

The emerging rules of thumb:

→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.

→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.

→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.

The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.

Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.

As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.

https://t.co/DNq60Skkmc