by Luis Rodrigues
What I’m building with AI + one hot take + four links worth your time.
The Take
I was watching the Google I/O keynote when Pichai dropped the number. Google is processing 3,2 quadrillion per month. Last year it was 480 trillion.
That's a 7x jump in 12 months.
They showed a lot of things: the Spark agent, new features across the board. But the real story for me is that those 3,2 quadrillion tokens per month aren't being generated in chat windows.
This growth is coming from AI running in the background, handling tasks in an inbox, executing workflows without the user writing a single prompt.
Google just showed everyone where software is moving.
AI won't be a feature users will click to open. It's an infrastructure that is always available and running in the background while they're doing something else.
This is the reason why Spark runs in a cloud VM and is not a tab. Same reason why Google has built agents into Ads, Analytics, and other properties instead of building a standalone chat assistant.
If you launched a product in the last two years with a chat interface glued to a set of features, you built something for an era that is fading.
Users won't open a chat to ask questions; they expect contextual data to appear when they need it.
Google is training 1 billion users to expect software to work without being asked. Your product is either ahead of that curve or behind it.
Together with
Every meeting I have about Thoughtled generates action items, and I always forget some. Now Granola runs in the background, no bot joining the call, transcribes everything. New users get the first month free → https://www.granola.ai
The Build Log
A thoughtled.ai client who focuses on cross-border delivery had a big announcement last week.
The marketing team drafted a post, and 4 employees published it in their own voice and got 7.5x more views than the company page's post.
That content wasn't rewritten 4 times. It was adapted from an original post using an agentic pipeline that applies a person's tone of voice to a piece of text.
This is a perfect example where AI excels today.
This week, I wrote a LinkedIn post that has hundreds of thousands of impressions. The hook was simple: people who actually use AI know where it breaks.
The post struck a nerve, but it was missing something. It didn't answer the question of where AI is really useful today.
There are three areas I see real value being created.
Corporate knowledge: every company has valuable content buried across Notion, Drive, Slack, SharePoint, and more. GenAI enables people to search in all those sources and generate an answer based on real documents.
The difference between "search" and "find".
Customer support routing: it does not replace the support team, but does initial triage and solves a lot of the simple queries. Intercom has reported an aggregate of 67% of questions answered by AI agents across their clients. A very large company here in the UAE resolves 50% of its customer support queries using AI agents in seconds, with a higher NPS than when humans handled them.
The AI answers what it can and routes the rest to human teams. Segment-of-one content: you write one original piece of content, and AI will adapt it for each of your clients. It's like having unlimited copywriters to adapt your message for each and every customer (not a segment, a real customer).
This is how thoughtled.ai works. Five-stage pipeline: clean, build a voice profile, rewrite, critique, revise. I'll break down the full architecture in a future article.
That's where AI fits. Where it doesn't: anything deterministic.
Never ask an LLM to calculate the average of a column of data. LLM is designed for language, not mathematics.
Need an average? Let the LLM write the code, then run it.
If you're not building with AI yet, here's something very easy to try. Put an AI router in your support emails, even something simple like n8n can save you hours repeating the same answers.
Or take a few articles/emails you wrote and ask Claude to rewrite a corporate press release in your voice. You'll be surprised how much it will sound like you.
On My Radar
762 points on HN. Google auto-updated its IDE into a completely different product overnight.
Open-weight models on commodity hardware are catching up to frontier APIs. If you're building, your unit economics just got a second option you didn't have six months ago.
Durable execution, isolated environments, self-healing infra. This is the gap between a demo agent and one that runs in production without babysitting.
$3B ARR for a code editor. This is proof that AI-assisted development is now an infrastructure that companies are willing to pay serious money for.
What are you building?
Got something you built? Hit reply, the best ones get featured next week.
Know someone who needs to write better content? Forward them this.
New here? Someone forwarded this to you? → Subscribe to Build What Matters: https://luisrodrigues.ai/
How was this issue?
P.S. Subscribers get $5K+ in software credits through our Secret partnership (AWS, Loom, Notion, Lovable, and more). Haven't claimed yours yet? → https://build-what-matters.joinsecret.com/
