Whats up whats uppp!
You're getting this email because you subscribed to aidstation, the newsletter where I share what I'm using, what I'm learning, and what's happening in AI. If you're new here, welcome.
Let's get into it.
What I've been using
ClawdBot
I've been playing with ClawdBot and I think this is what personal AI assistants should have been from the start.
The idea is simple: an AI assistant that runs 24/7 on your own server and talks to you through WhatsApp, Telegram, Slack, Discord, Signal, or iMessage.
Not a website you visit. Not an app you open. It lives in your messaging apps, like texting a really smart friend who never sleeps.
Siri still can't remember what you told it yesterday. ClawdBot does.
It’s pretty much what I’ve been trying to accomplish with Claude Code for the past months.
Mention you have a meeting on Friday, it remembers. Tell it your preferences once, it keeps them. The more you use it, the more it knows about how you work.
But the big difference is that it's proactive.
Most AI assistants wait for you to ask something. ClawdBot reaches out to you. Morning briefings with weather, your calendar, reminders. Alerts when something you care about happens.
One guy on Twitter e has it timeblocking tasks in his calendar based on importance, scoring tasks with a custom algorithm they developed together, and leading weekly reviews based on meeting transcriptions.
And it can actually do things. Browse the web, fill forms, read and write files, run shell commands, check your email, send messages.
It has 50+ integrations out of the box: Spotify, GitHub, Gmail, Twitter, Obsidian, and more.
And if something doesn't exist, it can write its own skills. Someone on Twitter said they wanted to automate Todoist tasks, and ClawdBot created the integration itself during the conversation.
The setup is technical. You need a VPS (around $5/month on Hetzner), Node.js, and to run through an onboarding wizard. It takes maybe an hour if you're comfortable with a terminal. Plus you can also ask Claude Code to help you set it up
The repo is open source at https://clawd.bot/ and there's an active Discord if you get stuck.
The cost: around $5/month for the server, plus whatever you pay for Claude. Pro at $20/month works, Max at $100-200/month if you want more.
My setup right now: I have it on Telegram doing morning summaries, tracking GitHub activity on our repos, and reminding me about follow-ups I tend to forget. Still experimenting with what else to automate.
The honest take: the onboarding is rough. It's an open source project, not a polished product.
But once it's running, it feels like having a personal assistant that actually works. Not the "assistant" that Apple and Google have been promising for a decade.
If you set it up, reply and tell me what you're using it for. I want to steal ideas.
What caught my attention this week
ChatGPT is getting ads
OpenAI announced that ChatGPT will start showing ads in a few weeks. It starts in the US and only affects free users and the $8/month Go plan. The ads will appear at the bottom of answers, they say it won't affect the responses, and no conversation data gets shared with advertisers.
What's interesting is the context. OpenAI made $20B in revenue in 2025, more than 3x what they made in 2024. Sounds like a lot until you compare it to Meta ($180B+) and Google ($295B), and that's just from ads. OpenAI is still tiny in ad revenue terms, but they clearly see the opportunity.
My take: if you're using AI for anything serious, you're probably paying for it anyway. But for the millions of casual users, ads were inevitable. The question is how intrusive they get over time.
Context Engineering is the new Prompt Engineering
We've been talking about "prompt engineering" for years, but the real skill now is "context engineering", the art of filling the context window with exactly the right information for the task.
The problem is something called "context rot." As the context window grows, model performance actually degrades. There's apparently a "pre-rot threshold" around 128K-200K tokens, way below the 1M+ limits that companies advertise.
The four pillars according to recent research: Writing (save info outside the context), Selecting (pull only what's relevant), Compressing (summarize, remove fluff), and Isolating (split context across multiple agents).
This explains why just dumping everything into the prompt doesn't work. The skill is knowing what to include and what to leave out.
AI isn't 10xing developers. It's 10xing PMs.
This take from Steve at Builder.io made me think.
His argument: PMs live in fog. Imperfect data, shifting markets, endless debates, long bets that age poorly. The cure is speed. Real MVPs in real users' hands. Fast loops, not long plans.
With AI, PMs can now go from feedback to prototype to validation without waiting on anyone. Engineers focus on the hard parts. Designers refine in place.
The punchline: "AI hasn't made code 10x faster. It has made clarity 10x faster."
I think there's something to this. The bottleneck in most projects isn't writing code, it's figuring out what to build. AI is compressing that part.
Nobody wants to hire juniors anymore
A Harvard study tracking 285,000 firms found that junior roles are down 23% at companies using AI. Senior roles are up 14%.
The math: Before AI, 1 senior + 3 juniors = 4 person team. After AI, 1 senior + Claude = same output.
But here's what nobody's talking about: if you don't hire juniors, where do seniors come from?
The CEO of Zoho put it well: "AI makes senior architects more productive and reduces the need for junior engineers. But if we don't have junior engineers, we don't get to train the next generation of architects."
We're eating our seed corn.
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.