Whats up whats uppp!
Why you're getting this: I'm sharing some tools and approaches we use at Sidetool and personally. I think you'll find them useful.
Zero pressure to stick around, just click unsubscribe if this isn't for you.
Let's get into it.
What I've been using
Custom Funnel System (bye Perspective)
We used Perspective for our marketing funnels. It worked really well for tracking the whole user journey, integrations with Slack, CRM, analytics.
Building each landing was painful. You're stuck with their templates, every change is manual in their editor, customization has limits, and you don't own your data.
With Claude Code that changed completely.
Now I just give it the site repo, explain the message and structure I want, and in ~30 minutes the landing is done. Because we already have our own templates, every new funnel compounds. Less effort, more speed, and tighter iteration each time.
The thing is, if you build landings with code, you lose all the tracking that Perspective gave you for free.
So we built our own system. A tracking script that captures the whole journey:
Where the user comes from (UTMs, referrer, device type)
Every step they view and complete
Every answer they give to funnel questions
Partial form data if they abandon mid-way (email, name, phone saved as they type)
The event they schedule
Final conversion with all data attached
The script generates a unique device ID stored in localStorage, so we can track the full session even if they leave and come back. It auto-captures UTMs from the URL and sends everything to our API.
The data flows like this: FunnelTracker.js calls /api/track on Vercel, which has the Supabase key as an environment variable (so nothing sensitive is exposed on the landing page), and writes directly to the database.
In Supabase we have 4 tables: funnel_visitors (each unique visit), funnel_step_completions (each step completed), funnel_conversions (leads with full data), and funnel_registry (funnel configs).
When someone converts, a separate serverless function hits Slack and our CRM automatically.
The architecture follows an event-driven pattern: Trigger Event → Store Event → Subscribe to Event.
The script triggers the event, the API stores it in Supabase, and then you can create multiple reactions from the database.
Want to send a Slack message? Subscribe to the conversion event.
Want to create a lead in your CRM? Another subscription.
Want to send a welcome email? Same pattern.
Each reaction is independent, so you can add or remove them without touching the tracking code.
The combo of Claude Code to build fast plus our own tracking system is brutal. We have total control of the data, we can run SQL queries directly, build custom dashboards, A/B test with a simple parameter in the URL (?variant=b).
Oh, and we saved the $350/month we were paying. That's $4,200 a year.
If you want to see the tracking script and architecture breakdown, reply and I'll send it.
What caught my attention this week
This is how Claude Code is meant to be used
Boris Cherny, the guy who created Claude Code at Anthropic, posted a thread showing how he actually uses it. It went viral.
His setup: 5 Claudes running in parallel in his terminal. 5-10 more in the browser. Sessions from his phone. All at the same time.
The result: 259 PRs in 30 days. 497 commits. All written by Claude Code with Opus 4.5. "Last month was my first month as an engineer that I didn't open an IDE at all."
The most interesting part isn't the numbers, it's the workflow. He starts every session in Plan mode. Goes back and forth with Claude until he likes the plan. Then switches to auto-accept and Claude usually nails it in one shot.
His team shares a single CLAUDE.md file in their repo. Every time Claude does something wrong, they add it to the file so it doesn't happen again. It's like building institutional memory for your AI.
I've been experimenting with taking this a step further: self-improving skills that automatically analyze your session, extract corrections, and update themselves. Correct once, never again. If you want to see how I'm setting this up, reply and I'll share.
The key takeaway: "Give Claude a way to verify its work. If Claude has that feedback loop, it will 2-3x the quality of the final result."
A Google engineer said Claude Code built in 1 hour what her team spent a year on
Jaana Dogan, Principal Engineer on Google's Gemini API team, posted this and got 5.4 million views in hours.
She gave Claude a 3-paragraph description with no proprietary details. And the result wasn't perfect, but it was comparable.
The interesting part was what she said after: "It takes years to learn and ground ideas in products, then come up with patterns that last for a long time. Once you have that insight and knowledge, building isn't that hard anymore."
That's the shift. The hard part is knowing what to build. Once you know, building is trivial now.
OpenAI is building an AI pen with Jony Ive
OpenAI and Jony Ive, the guy who designed the iPhone at Apple, are building a pen-shaped AI device. Codename: Gumdrop.
It has a camera that watches you write, digitizes your handwriting instantly, and uploads everything to ChatGPT. You can also have voice conversations with it. It runs OpenAI models locally.
The interesting part is the philosophy. They're calling it "simpler than an iPhone." The idea is focused interaction with AI, not another screen competing for your attention. Ive and Altman have emphasized a "calmer vibe" in interviews.
OpenAI put $1 billion into the partnership and bought Ive's startup for $6.5 billion. Foxconn is manufacturing it in Vietnam. Expected launch: late 2026 or early 2027.
They're also rebuilding their audio model from scratch for Q1 2026, more natural speech, faster responses, better interruption handling. The pen is clearly part of a bigger bet on voice-first AI.
Waymo's safety numbers are insane
Alex Immerman from a16z wrote a piece on self-driving cars this week that had some wild stats.
Waymo's data over 127 million driverless miles: 90% fewer crashes causing serious injury. 92% fewer crashes with pedestrians. 83% fewer with cyclists.
His take: "If a mayor blocks deployment of a system that cuts serious crashes by 80-90%, and people keep dying on those roads the old way, that should be treated as a moral crime."
The interesting framing: autonomous vehicles are "social robots." They learned to have visual conversations with other agents on the road, yielding, passing, reading hand signals. It's like conversational AI but through movement.
You can check out the full post here.
That's it for this week.
If you try any of these, or have thoughts on what's happening, reply. I read everything.
-Ed
Did you enjoy this newsletter? Say thanks by checking out one of my businesses:
Liked this? Sign up here to get more.