I am working to automate my job 🤖
Anything that an agent can do, an agent should do it in my place 👀
Turns out, mostly anything that has to do with a machine can be automated. Here are a few examples 👇
Agent can handle it all
The key first step is noticing 💡
Assume that an agent can handle all the tasks 🙌
A few of those tasks are more critical than others. That's on you, and you are responsible for it 🫵
But many smaller/intermediate tasks don't need your constant presence. Not at all. Your agent should have a way to handle all of it.
Make an habit to notice all the tasks that are less critical, and how an agent could do it instead 🤔
This is not (yet) about engineering or automation. It's just noticing: being aware that agents can do most of it, and note when you are still handling something you shouldn't.
Engineering automation
Now you have a list of tasks you want to automate.
There are multiple levels of "automation". What I am aiming for is complete automation:
The agent does not need my presence or even my prompt, it just does things 🪄
Example: requests from the support team 👀
I noticed over the years that most requests and bug reports can be answered with a simple codebase lookup. And an agent can do it 💁🏼♂️
So, I did the following:
- Create a Slack bot/app
- Connect Slack webhook to Grok Bot (AI on the cloud)
- The support team asks questions to my
SandroBot SandroBothas access to Github, Linear, server logs and more
Many (many!) of the requests are answered directly by the bot (without me even noticing).
A few are real bugs: the bot will then escalate to me to take a look (using Linear to create a detailed task) 🪄
Access and information
Turns out that a lot of a software engineer day-to-day work was collecting information 💁🏼♂️
In fact, if you give your agent access to those information, you are pushed out of the loop
Here are some examples:
- Github access with context of codebases and associated deployments
- Access to telemetry, logs, deployment status
- Access to your production app with a testing account
- Access to Slack with all the team messages context
- Access to analytics events (Sentry, PostHog etc.)
- Database (with proper permissions)
- Cloud deployment
Anything that has to do with checking the status of the app, explaining features, inspecting UI, QA, and more, all handled by the agent, with proper access.
What's left on you? 🤔
Quite a lot still. For once, setting up and managing the agent automation system 💁🏼♂️
Second: responsibility.
No matter what, someone has to manage and direct the system. Maybe fewer people, but you are still critical to make sure it all goes smooth.
Or clean up the mess when needed 💁🏼♂️
See you next 👋