Howdy wizards,
I’m having lots of fun visualising the landscape of AI use cases lately. If you have a special interest — reply and let me know what you want to see next, and I’ll see what I can do!
Here’s what’s brewing in AI.
What’s the highest-impact AI use case in your domain?
Every business function has different areas which are better suited for AI than others.
Not every process gains real economic value from being automated or improved with AI.
Within each domain, here are the use cases that score the highest on Context Windows’ Proven Impact score.
Based on 2,145 AI implementations from the last 24 months
The standout things:
Predictive maintenance in operations is a killer use case. The studies where companies bought a solution versus built a custom one have notably better outcomes.
Example: Georgia Tech used AI to analyse real-time sensor data from solar panels and irrigation systems to optimise energy, and saved $2.3M.
Lead qualification is an easy, massively effective use case for sales. I covered this in detail last week.
Content creation for marketing is the entry on this list where build vs buy does not change the proven impact score meaningfully. Content workflows are often highly individual/hard to standardise in a tool — but straightforward to automate when you already have lots of experience doing it manually.
Example: Kraft Heinz used Google’s suite of tools to build TasteMaker, an internal platform that lets the marketing team visualise on-brand product concepts and test them against consumer research data.
High on this list is also customer service + conversation intelligence. These are tools that evaluate all the unstructured data of spoken conversations — and score QA, analyses sentiment, flags competitor mentions, and more.
Example: MSC Industrial Supply went from analysing 36,000 calls per year to 2.5 million. An AI scores every customer support call and allows them to handle things more proactively.
—
My product Context Windows runs on open web data, which I turn into structured intelligence about how companies are using AI.
For that to be useful, I need the best coverage, timeliness, and accuracy on what's being published.
That's why I've teamed up with CatchAll by NewsCatcher — an AI-native web search API.
I use it to track every company's AI implementation I can find across the open web. But the use cases are wide open — feed it any query, get a clean structured dataset back.
IN PARTNERSHIP WITH NEWSCATCHER
To build a great dataset from web data, I have 2 rules:
#1 is to not use Deep Research. Unless you like randomly sampled data with a sprinkle of hallucinations.
#2 is to use CatchAll, which scans literally thousands of pages, and gives you accurate, clean data you can use right away.
Every cybersecurity incident this week. All funding rounds in a sector. Every regulatory action affecting your industry.
These aren't questions with a single answer — they're datasets waiting to be built.
CatchAll is an API that turns queries like these into structured, deduplicated records — extracted from across the open web and ready to use.
Turn any query into a live monitor and keep it up to date automatically.
NEWS NEWS NEWS ❦ NEWS NEWS NEWS
The big things
1
Salesforce unveiled Headless 360, exposing its entire stack to agents via APIs, MCP tools, and CLI commands. Agents can now drive Salesforce data and workflows directly without a browser UI.
Why it matters
For a company whose value prop was the dashboard, this is a significant admission: the future primary user of enterprise SaaS isn't a coffee drinker, it's an agent calling functions. Might have to adjust that price-per-seat strategy too.
2
Cursor signed a strange one-year option with SpaceX/xAI, and is also raising $2B at a $50B valuation. The SpaceX side works like this: xAI gives Cursor Colossus-scale GPU access to keep training its in-house Composer model, and in return SpaceX earns the right to acquire Cursor for $60B by year-end, or pay $10B to walk away. I know, that’s a mouthful, why’d they have to make things so complex. Anyway…
Why it matters
The less obvious story here is that Cursor might actually need this to survive long-term. On the surface, Cursor is a crazy growth story: 3 years in they already have $2B ARR and 1M+ paying users. But most Cursor usage today still routes through Claude and Codex, and both Anthropic and OpenAI are aggressively building coding platforms themselves to keep users within their ecosystem, and also blocking out 3rd party tools. Right now Cursor is too dependent on those suppliers, especially considering the conflict of interest. So this would be a type of vertical integration that would give Cursor more of a moat and also xAI an actual use case.
All the small things
Industry moves
Tim Cook is stepping down as Apple CEO on September 1 after 15 years, with hardware chief John Ternus taking over. Cook moves to executive chairman, leaving a $4T company with a much bigger services and wearables footprint but an AI strategy that's still unresolved. Ternus is a career Apple hardware engineer; that’s telling about where Apple thinks their moat is in this AI era.
Anthropic CPO Mike Krieger resigned from Figma's board on April 14, three days before Claude Design launched. Clear signal that AI labs plan to compete with the SaaS layer.
Meta will start logging employee keystrokes, mouse movements, and periodic screenshots to train AI agents. The tracking runs in VSCode, Metamate, Google Chat, and Gmail. No opt-outs allowed, but US-only because EU has laws and stuff. Also, they’re cutting 8,000 jobs (10% of their headcount) next month, and the departing staff will have their workflows recorded for a month on the way out. Yikes.
Snap is also on a cutting spree: around 1,000 jobs (16% of its workforce) in the name of AI productivity and profitability. Snap projects Q1 rev will be up 12% — this isn’t about revenue. It’s the "AI is making us faster, so now we need fewer people" narrative that every large-cap tech company has secretly agreed on for this quarter.
GitHub paused new Copilot signups as usage and cost pressures force a switch to token-based billing and tighter rate limits. They also removed Opus from pro plans. This is a clear tell that subscription models are under huge pressure with all our agent workflows. Even at incumbent scale, these unit economics aren't holding.
Cold brew
❦
Wishing you a solid day with lots of ice cream.
You are a delight.
Dario
See which AI use cases are paying off
Most companies pick & prioritise AI use cases by brainstorming internally + guesswork.
A moment of silence for the 90% of those initiatives that fail
I’ve spent months building a system that lets you validate your ideas against what 2,000+ companies have already succeeded with.
🟦 Start at contextwindows.ai or book a demo with me
What's your verdict on today's email?
Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.