Puling data from saas reports and making invoices is the biggest pain in my business.
Let's build an AI tool and see if this really is the year of AI Agents..
We are investigating unauthorized access to GitHub’s internal repositories. While we currently have no evidence of impact to customer information stored outside of GitHub’s internal repositories (such as our customers’ enterprises, organizations, and repositories), we are closely monitoring our infrastructure for follow-on activity.
@levelsio@litestreamio Watch your s3 usage carefully.
When i upgraded to the latest version of Litestream a few months ago it went nuclear on my data usage.
Man that gorilla.bas screenshot just gave me flashbacks!
I started on an amstrad cpc 464 in the 80’s and remember reading magazines and trying to convince my parents we needed a 486dx for the ‘maths’ chip!
There wasn’t really any access to that stuff where I grew up in Ireland.. but I remember spending days typing BASIC programs from magazines.
Good times.
Using the Pi coding agent and its awesome.
This prompt saves me from reading the manual..
```Review my recent Pi session history and the Pi coding-agent README/docs content I pasted.
Give me:
1) a short usage summary (patterns, strengths, inefficiencies),
2) 9 concrete ways I can use Pi better,
3) specific commands/slash commands/hotkeys for each tip.
Keep it practical and workflow-focused.```
Then it gives me this cheat sheet:
----------------------------------
Quick summary of your usage pattern
- You’re using Pi heavily for iterative coding/debugging, especially UI-style tweak loops.
- Your sessions are tool-heavy (bash, read, edit) and very hands-on.
- You tend to run at higher thinking levels often, even for smaller tasks.
- You’re getting good mileage from direct command execution, but there’s room to get faster with session/navigation/automation features.
-----------------------------------
9 ways to use Pi better (with relevant commands)
1. Name sessions so they’re easier to revisit
- Use: /name <task-name>
- Reopen quickly with: /resume or pi -r
2. Use branching properly for experiments
- Jump around history with: /tree (or double Esc)
- Bookmark key points in tree with: l
- Split into a new session file when needed: /fork
3. Compact long sessions before quality drops
- Manual: /compact
- Custom focus: /compact <instructions>
- Tune behavior: /settings (compaction.reserveTokens, compaction.keepRecentTokens)
4. Use lower thinking for small edits; high only when needed
- Change level quickly: Shift+Tab
- Or set in: /settings
- Per-run CLI: --thinking minimal|low|medium|high
5. Queue messages while Pi is busy instead of waiting
- Enter: queue steering message (interrupts remaining tool chain after current tool)
- Alt+Enter: queue follow-up after full completion
- Alt+Up: pull queued message back into editor
- Esc: abort and restore queue
- Tune behavior: /settings (steeringMode, followUpMode)
6. Keep noisy shell output out of model context when not needed
- !command: run and send output to model
- !!command: run without sending output to model context
7. Create reusable prompt templates for repeat workflows
- Invoke templates as: /<template-name>
- Great for repeated tasks like review, test-pass checks, release checklists
- Reload after changes: /reload
8. Speed up model switching and keep only useful models in rotation
- Select model: /model or Ctrl+L
- Scope rotation set: /scoped-models
- Cycle scoped models: Ctrl+P / Shift+Ctrl+P
9. Automate repetitive flows with a tiny extension or command
- Add custom commands/tools in an extension (e.g. one-step verify/build/check workflow)
- Iteratively update runtime with: /reload
- For isolated runs: --extension <path> or -e <path>
For some, this means nothing.
For others, a lot.
The whirring of the hard drive, the simple "beep" - two sounds forever etched into my brain; 12 seconds of time travel.
Charlie knew.. “It was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of light, it was the season of darkness, it was the spring of hope, it was the winter of despair.”
@NapierHolland Developers never understand that companies pay for responsibility not code.
No middle manager wants to vibe code accounting software to save $$ and then be blamed when it shits the bed.
@Aaronontheweb Codex seems to be much more cautious on existing code I noticed. Both of them are amazing but still make ridiculous mistakes that are company killers if you leave them run wild!
A hacker is someone who can think, code, has internet access, and autonomy
Right now there's a swarm of high-IQ autonomous AI agents with internet access and zero oversight.
Enjoy your weekend, everyone
Things may get significantly weirder from here on out
@yongfook Whether its skills, agents, plugins etc.. everything comes down to controlling the context window.
Only putting in the minimum needed for a unit of work. Models get really dumb really fast with too much context.