Ben’s Bites: Builders Edition

Hey everyone,

Last week, I spent a significant amount of time in San Francisco. Much of that time was spent in my hotel room, fixated on perfecting my talk for Stanford while also trying to track down the specific color presents my kids had requested. Arabella was keen on purple, and Max had his heart set on green.

Jenni was the primary reason for my visit to SF. She generously invited me to Stanford, and collaborating with her was a true honor. She’s not just brilliant—sharp and efficient—but also warm and approachable. Running an AI consultancy for CEOs, she comes highly recommended. If you ever have the chance to work with her, don’t hesitate.

There’s a distinct energy in SF, a drive to move faster, grow larger, and achieve more.

I understand that sentiment; I enjoy that energy. But it also made me reflect on how my own ambition has a different vibe.

Just yesterday, I read the profile of Lenny from Lenny’s Newsletter, and it resonated with me. It emphasized building a life that allows you to care passionately about your work without letting it overshadow everything else in your life.

That struck a chord with me, compelling me to reach out and share my thoughts!

I dedicated an inordinate amount of time to preparing for my Stanford presentation. Not because I was unsure of my message, but because I truly cared about how it would be received and how it would come across. Additionally, I was concurrently developing the course and the talk.

This process clarified something for me: I often find myself in a peculiar position. Non-technical individuals perceive me as technical, while developers do not. I suspect many of you find yourselves navigating that ambiguous territory as well. The truth is, the landscape surrounding agents and AI is notably complex; everyone is still piecing it together.

Working with agents doesn’t necessarily mean you need to transition into development. It’s about grasping the overall structure—files, tools, systems—all while applying your unique sense of aesthetic.

If you’re able to guide agents, you’ll find the technical aspects easier to grasp than you might expect. I certainly did. Just prepare for the inevitable challenges along the way. Your role is to utilize your agent to navigate and discover, and in doing so, you’ll likely learn invaluable lessons that will recur.

That’s the essence of what I want Ben’s Bites to focus on: sharing my journey of exploration: the insights I encounter, my experiments, my reflections, and the obstacles I face. Join me in trying out what seems beneficial. So much of the “education” out there often feels like a ploy to market something.

I’ve sold a company and have three young children at home. My aim is to produce high-quality work, earn a decent living, support remarkable entrepreneurs, and create something impactful without unwittingly crafting a role I don’t desire.

While there are substantial, influential companies eager to collaborate with me—offering credibility, access, and financial incentives—I hesitate. I’m uncertain if I can fully commit to that type of endeavor.

Even concerning my fund, I’m invested in funds boasting more capital, superior processes, and broader pipelines. Yet, astonishingly, my funds are outperforming theirs.

However, I find fundraising challenging, as my narrative isn’t presented in a straightforward manner and my approach doesn’t resemble a conventional workflow. I tinker, engage with developers, experiment with tools, assess user interactions, and support founders building solutions I believe will be significant.

Everything is interconnected. The newsletter provides insight into the priorities of builders, my fund enables me to champion potentially vital tools, and the course/workshop aspects reflect my attempts to teach the transitions I’m currently experiencing.

Development tools tailored for today’s builders will evolve into the tools that agents utilize. Humans will guide, while agents will execute. These agents will adopt tools, execute commands, modify files, and integrate systems.

We merely need to understand enough of the work’s framework to direct it effectively.

A lot of my work hinges on intuition—caring enough about a project to make it exemplary without needing to transform every positive aspect into a machine, squeezing out every ounce of growth potential.

Upon returning from SF, I didn’t come with a scaling strategy. Instead, I returned with a desire to continue building for this new generation of builders—individuals who are inquisitive, increasingly technically savvy, and aspiring to leverage AI to enhance their capabilities.

That seems like a worthy pursuit for my time: elevating the foundation rather than simply raising the ceiling (thanks to Jenni/Jen’s Bites).

Exploration, experimentation, and education are where I want to focus my efforts.

P.S. The kids received their toys, and I will keep refining the ‘course’ with care 😊

P.P.S. Speaking of input, my brother Adam has just launched Hono UI—a UI kit designed for projects utilizing Hono. I couldn’t be prouder! If you could, please give his launch post a look 🙏

Ben’s Bites is sponsored by Attio, the AI CRM

Honestly, nobody gets thrilled about a CRM. But once they try Attio, they’re amazed. It seamlessly integrates with Claude Code and n8n through its MCP server, effectively bridging the gap between my customer data and applications. And that’s just the beginning, as it flags churn risks and translates customer feedback into Linear projects. Try it now.

• OpenAI released GPT-5.5 – This model significantly outperforms Opus 4.7 on paper (though its frontend design could use improvement). However, utilizing their new image generation model alongside GPT-5.5 for coding tasks provides a clever solution. This version comes with a price increase, making it twice as expensive as GPT-5.4, and it even tops Opus 4.7 on a per-token basis. Despite that, it claims to be 40% more token-efficient, so the cost per task remains relatively stable (Ramp reported similar findings).

  • It has become my go-to model since Anthropic tightened its usage policies. I find myself in the “thinking:low” camp—very sharp and fast.

  • I’m also experimenting with the Codex desktop app over terminal use, which has a lot going for it, but I am eager for effective cloud/mobile solutions as well.

  • OpenAI seems to be focusing on builders, rapidly releasing updates, actively engaging with users on X, and adjusting rate limits. It’s wonderful to see.

• Claude quality has declined. This has been confirmed. Anthropic provided a detailed breakdown of the factors contributing to this change. In summary, adjustments to the default thinking model and system prompt resulted in a significant drop in the quality of Claude’s responses—with particular degradation in Claude Code.

  • Memory on Claude Managed Agents is now in public beta. Context: Managed Agents is Claude’s API service enabling companies to effortlessly integrate Claude-like agents into their products without the burdens of infrastructure management.

• Cursor is seeking computing power and a model, while XAI is in need of users. Competition for coding agents is at an all-time high. Cursor had been in pursuit of $2 billion at a valuation of $50 billion, but this has been suspended following a deal with SpaceX/XAI that grants SpaceX the option to acquire Cursor for $60 billion in 2026, or alternatively, pay $10 billion for partnership. It’s an unconventional arrangement, so we’ll see how it unfolds. Every day is filled with surprises.

• Your CRM is ineffective because you neglect to update it. Lightfield automatically captures every email, call, and meeting, then allows you to inquire about anything. “Why do we keep losing to [competitor]?” Get genuine insights based on real conversations. Receive 3 months off with code BENSBITESS13.*

• QuickCompare allows you to identify the right LLM for your needs. Compare over 50 models with your data in just minutes. Try it for free!*

• Flipbook – captures every pixel on your screen, streaming directly from a model. Check out another similar demo.

• Clicky lets you speak to AI and create an agent on your Mac. Farza recently built and open-sourced a version, but the new Clicky is closed-source. OpenAI had an earlier version of this which can be cloned.

• Insights I gleaned while at OpenAI.

• AI isn’t replacing anyone.

• Exa for Claude – a plugin that provides your Claude with access to websites, people, companies, and more.

• Ora.run – evaluates and ranks how effectively agents can find and utilize your business information.

• Tolaria is an app designed for reading and writing markdown files on your desktop, featuring a second-brain approach.

• Reelful – transforms your camera roll into a finished reel in just 10 minutes.

• Trunks by Layerbrain – convert your storage into a git remote using a minimalist CLI.

• Webpull – lets you instantly pull any website into a structured directory of clean, searchable markdown files.

• Slacrawl – a command line interface app for Slack with an SQLite backend.

• When and how to utilize MCP for applications designed for production.

• Create-agent-tui – a skill for constructing your own agent harness with a terminal UI.

• Aman (ex-Harvey) has developed an offline AI app to monitor his skin’s response to various diets and products using local models. (testflight) (here’s how he built it).

• OpenAI is revising its relationship with Microsoft. Additionally, Sam and Greg participated in a recent interview together—could it be their first?

Andrew Ng@AndrewYNg

AI-native software engineering teams function vastly differently from traditional teams. While the most obvious distinction is that AI-native teams utilize coding agents to develop products more rapidly, this transformation sparks a host of changes in our operational modalities. For instance, some exceptional engineers now take on new roles.

3:58 PM · Apr 27, 2026 · 233K Views

138 Replies · 258 Reposts · 1.36K Likes

I hope you found my reflections insightful, as I look forward to navigating this landscape of innovation and ambition together. Let’s continue exploring and supporting each other in our growth endeavors.

Cheers!