How to be human in an age of AI Agents
Content TBD
There is a schism in how I use LLMs at home and at work. I minimize their use when communicating with friends and creating and consuming art for pleasure and meaning, but I use LLMs profusely at work.
A large part of this website explains how I work, so this philosophy should be addressed.
“Work”, to me, is a bundle of tasks performed during contracted hours to advance an employer’s goals. It is a necessary part of life that one pursues to sustain their lifestyle.
These tasks can be decomposed into the following:
| Type | Produces | Examples |
|---|---|---|
| Production | The output your role exists for | Writing code, laying bricks, drawing shapes, talking to a client |
| Coordination | Alignment, so production lands | Pre-meetings, one-on-ones, feedback, decision memos, status updates, scheduling |
| Protection | Protection from future blame | ”Per our conversation” recaps, cc’ing your manager, written sign-off, approval screenshots, justification analysis, padded estimates |
| Maintenance | Keeps the system running. Office mise en place | MFA, tool switching, CRM cleanup, ticket tagging, report formatting |
| Improvement | Reducing the cost of future work | Building an automation, fixing an input form, writing a template |
| Learning | Expanding future capability | Learning a new tool, competitor research, trend reflection, mentoring |
In an effectively organized organization, protection time is zero, and coordination time is minimized. To me, they are externalities, with the former being most pernicious and indicative of a sinking ship that one should prioritize escaping.
Maintenance and Production tasks are necessary, visible parts of labor. Orgs sometimes prioritize production over maintenance, strategically or accidentally. Nobody gets rewarded at work for telling their managers how they prevented future costs.
Improvements and Learning have some of the highest returns on time. Its importance has varied by where I’ve worked. The best orgs explicitly budget time for improvement and learning. Even the most incompetent are eager to take advantage of individual learning and efficiency gains from expertise.
This is where AI tools (and most automation technologies) come in for the individual worker.
My Agents can take over protection and coordination completely. These are tasks for which every thought spent is a waste of precious life. Let your AI agent play Calendar Jenga and capture notes to make your micromanager happy.
Principles
- People before process before products. Start with how people think and work before reaching for automation.
- Primitive over packaged. Understand the seams where human judgment and machine execution meet.
- Deliberate delegation. Be explicit about what work belongs to agents and what remains irreducibly human.