A Snapshot of How I'm Using AI in January 2026

Avi Press | January 18, 2026

—

The way I've been using AI has changed dramatically in the last month. This post is a point-in-time snapshot of my personal AI setup for life and work right now. It will be outdated soon, and that's okay. It will serve as a fun reference point for later.

ChatGPT

I have ChatGPT Pro, mainly used for things like:

Voice mode for looking things up and learning about a topic interactively.
Deep research on a particular topic
Pulse is okay and sometimes has useful updates.

Clawdbot

Most of the work AI does for me is through Clawdbot running on a re-purposed old Macbook Pro in my home office, running on openai-codex/gpt-5.2. I message it from my phone or computer on Telegram mostly and sometimes on Slack. Telegram voice notes are used heavily when I'm on the go. Talking to Codex is dryer than a pack of Saltines but it gets stuff done and it does it accurately.

Context from my whole life can connect to Clawdbot

Everything running on my laptop means I can more safely* connect other systems. So far, I've given clawd access to:

Email
Calendar
Telegram
Slack
Linear
Metabase
GitHub
X
Ad-hoc site logins as needed (Amazon, auto-insurance provider, etc.)

Cron jobs

Clawd supports cron jobs, which is very handy. I have a few scheduled:

Every morning, check my calendar, linear and provide a morning digest.
Every morning, look through gmail and find customer / prospect conversations I need to follow up on.
Every morning, send a digest of the latest Scarf signups to our #sales channel, categorized by sales fit level and company enrichment.

Task examples from the last week

I'll often queue up tasks by spawning subagents. I can have multiple things going in the background without blocking the main interaction loop.

I had a cracked windshield and today clawdbot created the insurance claim on its own and all but grabbed the first available repair window, waiting for my confirmation.
Travel logistics and research for an upcoming vacation.
Generated a logo walls for a board meeting deck given a list of company names.
Asked ad-hoc data queries in Slack, in English, eg "How many new orgs were created last week?"
Fixed some bugs in Scarf's SDKs and sent PRs.
Edited a customer's custom Scarf report for the month. Clawdbot ran the query in Metabase, got the resulting CSV and uploaded it to S3. It then sent it to the #analytics channel in Slack for customer hand-off.

Surprises with Clawdbot so far

It's surprisingly robust and self aware.

Clawdbot largely moved itself from my daily use macbook over to my old macbook, which is now acting as a dedicated server in my home office. It ssh-ed into my old laptop and largely set itself up, needing me mainly for confirmations and some clarification.

Heartbeats are useful.

Clawdbot will check in from time to time via a heartbeat. Sometimes it's annoying, but more often than not it's following up on info or clarification that it needs, which is very helpful for pushing your task forward!

Things still break a lot.

Sometimes it just forgets it has credentials that it has. Sometimes other things just randomly break, like rate limits, model connectivity, or harness bugs.

It will gladly SSH into other machines if you want it to.

AI fleet coming soon.