A Snapshot of How I'm Using AI in June 2026
Avi Press | June 1, 2026
—
This is a follow-up to my previous notes on how I am using AI day to day and what my setup looks like.
Models
I'm using gpt-5.5 for nearly everything. I try the other models every few weeks but so far nothing has been compelling enough to endure switching costs of any kind. I use Pro-thinking from time to time.
Harnesses
OpenClaw is still the central system. I'm scheduling more cron jobs over time as the workflows have stabilized significantly in the last month or so.
I'm also using the Codex desktop app significantly more for more in-depth coding tasks. Codex also has some of its own automations, like pull request review sweeps.
Hardware
I'm still running OpenClaw on my old macbook, and it's still working great. I've abandoned Dropbox as a syncing mechanism for a shared drive of files between the machines, in favor of SyncThing + Tailscale.
Context and other orchestration
Org2
My side project Org2 is becoming central to my workflow, just as org-mode in Emacs has been for years. My daily notes, meeting recordings, emails, and more are all flowing into my org notes. Org2's structure and backlinking seems to work well for the LLMs to have good context from my work and life, and the new emacs-free CLI provides a deterministic API for things like agenda planning, encryption, backlinking, formatting, and more.
By staying largely compatible with existing org tools, I still have a reliable way for OpenClaw to trigger reminders on any device, and manage the todo list I actually follow day to day from my phone, laptop, but now from any editor.
I fun tangent: I've also multi-recipient encryption support to `org2-crypt`, which allows me to securely share secrets with individual agents via PGP. It's become a very ergonomic way to share secrets with OpenClaw that are encrypted end to end.
Development of org2 is also itself a very automated process at this point. I just ask OpenClaw to implement features or fix bugs as I come across them organically. I'm not reviewing its code at all anymore.
Some workflows
A few automations to give a sense of what's actual work is getting done:
Semi-automated software development
OpenClaw is continuously ingesting all of my meeting transcripts and notes, and opens Linear tickets where appropriate.
When a particular label is added to any Linear ticket, OpenClaw will begin working and send a PR. We are finding more places where this label can be added automatically. For frontend tickets, it will even post screenshots of the rendered UI changes for easier review.
Codex reviews OpenClaw's work continuously. When Codex signs off, OpenClaw will merge.
Deploys are still done by a human (for now).
User support
OpenClaw reads support tickets and drafts responses, and sends them after manual approval. We're increasingly giving it more access to actually respond to customer support tasks on its own.
Documentation updates
OpenClaw reads the git log of our production git branch, drafts a changelog, and opens a PR. It monitors for feedback until it gets merged.
Scarf AI review and improvement
OpenClaw reads through Scarf AI chats, and monitors for bad responses from Scarf. It sends a digest to us in Slack and follows up with Linear tickets where appropriate (which will then get implemented automatically once the relevant label is added).
Biggest open problems
OpenClaw still can't order my groceries for me. Amazon is still too hard for OpenClaw to reliably navigate.
Not enough online stores have implemented things like Link by Stripe for you to be able to thoughtlessly just ask OpenClaw to buy you something and have it Just Work.