Hey there!
Summary of this email:
- Understand how coding agent costs work (+ AI for Programming course 100% published)
- 💰 Get up to dIAte with €50 off
- 🗞️ Opus 4.7 launch is imminent!
- Joke of the week (this one's new!)
🤖 Understand how coding agent costs work
One curious thing is that if you say "Hello" to a coding agent, that first prompt can already consume almost 20k tokens.
That's because tokens, and therefore cost, aren't just about what you type into the prompt.
Each request carries much more context than it seems: the system prompt, the available tools, the skills, the memory files...
In other words, you send a short sentence, but the agent may be sending thousands of extra tokens before it even starts responding.
The message you send counts as input tokens. What the agent replies with counts as output tokens.
Output tokens are significantly more expensive than input tokens.
And since LLM APIs are stateless, every new interaction includes all the previous ones again. Your next prompt doesn't start from scratch. It starts with everything you've already said.
That means output tokens turn into input tokens on the next turn.
Understanding these two concepts, and how agent caching works, is really useful because it shows that there's room to cut costs.
Once you see that, it's clear that having a good AGENTS.md and good documentation is key to keeping costs down.
We explain all of this in the course Agentic Programming with AI: Practical Fundamentals, where we also go deeper into caching and how to get more out of agents while keeping costs under control.
💰 Get up to dIAte with €50 off
More than two years ago, we decided to go all in on AI.
Things are taking shape, and we already have plenty of material about it on both YouTube and Codely Pro.
From how to code with AI following good practices, to running models locally, to building your own MCP server.
Many of these courses are in the Premium plan because they required extra effort to prepare, while still keeping that Codely stamp of going one step beyond the basics.
We know the Premium plan has a considerable cost, so if you were waiting for a sign to finally catch up with everything happening in AI, here it is:
€50 off and Agent Harness Bootstrap included if you sign up for the annual plan this week.
🗞️ Opus 4.7 launch is imminent!
There are lots of rumors that Opus 4.7 will launch today or tomorrow.
Honestly, we're really looking forward to it launching already, so Claude Code can feel stable again, because it's been pretty nerfed over the last few days.
If it launches today, we'll try it and tell you all about it tomorrow on Café con Codely. If it doesn't, we'll tell you what our expectations are.
Tomorrow live at 9 CET, covering this news and more. Join us on YouTube or Twitch!
And since you've made it to this part of the newsletter, here's the joke of the week, which I know you were waiting for:
If I write exceptional code... is that a good thing or a bad thing? 😂 😂 😂
Cheers!