I'm on the claude code train

create Midjourney videos and automate browsers without coding

Jun 19, 2025

I write a newsletter about startups and investing—for ai builders of all levels.

I record mini-tutorials, review tools I’m testing, share my insights from an exited founder turned investor.

Hey folks,

I’ve been busy this morning…I’ve built;

a chrome extension to quickly navigate to the last tab i was on
a feed of all accelerator events (applications/demo days)
some tracking systems for us to find great products and news for this newsletter
a claude code clone for less-technical people

How?

Using Claude Code. It’s just phenomenal. Ian did a little intro for beginners but I think there’s more to cover so I may make one today/tomorrow.

You can use it directly in your terminal (eek, scary…until its not) or within the terminal in Cursor - and maybe soon in my version that i’m building.

Why?

I love making stuff, playing with these tools and learning about it all - the process, thought patterns and yes, how to deal with issues/bugs.

I also may have a chip on my shoulder because I was arguably the world’s no-code expert back in 2019 onwards and I’ve been at the forefront of all these capabilities, yet I didn’t solo-found a vibe coding company that sold for $80M in less than a year. Base44 did.

I’m not playing the building game any more. Instead I’m much happier tinkering and investing in people that are!

Speaking of builders, you need to watch the YC AI startup school presentation from Andrej Karpathy - which they graciously edited and uploaded very quickly (else it may be out of date!)

It essentially talks about the new software stack, how AI changes how we’ve previously done stuff and validates what I mentioned on Tuesday’s post - AI is a new computer.

🔎 News worth knowing

Gemini 2.5 Pro is finally out of preview, and they have a new Gemini 2.5 Flash Lite model. Practically, this new model is the replacement for the old Gemini 2.0 Flash—it has similar pricing with an optional reasoning mode. The 2.5 Flash model is better but relatively much more expensive. I know confusing, but I’m happy that I don’t need to change the model slug for 2.5 Pro every 2 weeks now.
OpenAI is doing it all. They have a new official podcast, Greg Brockman went on one with Stripe’s cofounder John Collison, you can now bring ChatGPT to your meetings if you’re on Pro, Enterprise, or Edu plans, too (sorry Plus plan users). And oh, they have new research claiming that models that write bad code turn out to be suspicious in other use cases as well, but that behaviour is fixable. They call it emergent misalignment.
Midjourney released their first video model, and as you'd imagine with Midjourney, it is insane at aesthetics. It is a bit limiting—you can only create a video from an image (your own or midjourney generated), animating the elements on that image in two modes, automatic and manual. But it's fast and cheap, bundled in the $10 midjourney subscription. I feel it's really great to create b-rolls and entry shots for your videos. More technical details here.
Jam.dev launched the first magic link for bug reports. It's a link you can send to your customers, and get back a screen recording with console logs and network requests. Resolve issues 2.7 days sooner. Skip the troubleshooting calls. Send a Jam link instead. Try it free.*

*sponsored

want to partner with us? Click here

🌐 What I’m consuming

Does MCP kill vector search - why do you need to store embeddings if you can get real time data on demand.
How to deploy a remote MCP server on Google Cloud, or a personal one on CloudFlare. From a community member, and super useful tutorial!
Difference between search for humans vs for AI.
Pope Leo takes on AI as a potential threat to humanity. The piece has less about the new pope but it has a quick coverage of the interaction between tech companies and the Vatican over the last decade.
What Google translate can tell us about vibecoding.
The practical guide to onboard Claude Code to your team (+ some keyboard shortcuts for Claude Code) or if you prefer a video:

⚙️ Tools I’m looking into

Yupp AI - This app lets you use all AI models for free by giving you credits in exchange for your feedback. For now, they have raised $33M to afford doing this, and the data from putting LLMs head to head can bring more $$, but is it endless? Like the ad revenue, to keep offering these models for free for a long time.
Proactor - AI agent that listens to your conversations in real time, proactively identifies needs, and acts before you need to ask.
Director by BrowserBase - Use natural language to create code for automating browsing-dependent tasks. Native AI tools (ChatGPT/Claude) really struggle at creating robust code for browser automations, so I’m looking forward to how this works.
Typedef - the AI data engine for modern workloads.
Dream Recorder - This open-source device plays your dreams back at you in fuzzy resolutions. On the same beat, Walter by Autograph works as your AI historian to remember your life stories.
remote-mcp comprehensive list of integrations that can be used as MCP Remote servers.
Firestarter - an open-source chatbot building platform to chat with websites. (repo)
Net - Email done smartly from your phone. fewer notifications, more control.
This fun side tool from Mindy turns your slack chats into memes.
You can now monitor new matches for your Exa Websets with improved data export/import support.

🥣 dev dish

Prompts are now primitives in OpenAI’s APIs.
You can now connect to remote MCP servers from Claude Code.
rulebook - universal rules template for AI coding assistants.
CloneDex - an open-source OpenAI Codex clone.
OpenHands CLI - open source command line tool with similar performance as claude code, open to all AI models.
Jan from Cloudflare stripped auth to a simple solution that’s easily extendible and understable with LLMs.
ai terminal for windows - turns your requests into powershell commands and runs them.
OpenAI released this demo of a Customer Service Agent interface built on top of the OpenAI Agents SDK. It’s a really good example if you want inspiration for building apps on the “humans will manage agents” premise.

🍦 Afters

Generalist AI is building robots that can do any task, not just specific tasks with a tema of ex-Deepmind and OpenAI researchers.
Multiplier Holdings - a technology company building AI-native professional services firms.
Headstarter is organising two batches of Buildcore (one in July and August) - a completely free two-week sprint for seasoned engineers to work on real projects from series A companies. Apply here.
Uncork Capital is hiring a Technical Associate - Someone with strong technical background to hunt, evaluate and support seed-stage companies.
Granola is hosting a chat with Anthropic London’s head of enterprise & startup sales on June 30th.
cursor is eyeing new funding at a reported $18B valuation—double of the latest announcement made just this month.
openaifiles.org - a compilation of all reports of OpenAI’s misdeeds. Nothing new here, a collection of what’s already been said.