Teach it about you once

Count how many times this course has had Priya type "I run operations for a logistics company in Pune." Every new chat starts from zero (lesson 2), so every new chat pays the same context tax — small each time, real in aggregate, and worse than the typing is what happens when you skip it: generic answers, because the intern forgot who you are again. As of mid-2026 every major assistant has machinery for paying this tax once. This lesson is about setting it up deliberately instead of letting it accumulate by accident.

The standing brief

The simplest machinery goes by names like custom instructions, preferences, or profile: a block of text the assistant silently reads at the start of every conversation. Write yours in two parts — who you are, and how you like answers:

About me: I head operations for a logistics company in Pune — about
140 vehicles, 60 staff, clients across Maharashtra. My work is vendor
management, fleet ops, compliance, and reporting to leadership.

How to answer: Be direct and concise. Use bullet points for anything
with more than three parts. Currency in INR unless I say otherwise.
When I ask for a draft, give me the draft first, commentary after —
if I want options, I'll ask. If something I've said seems wrong,
say so plainly.

Five minutes to write, and every future chat starts warm. The "how to answer" half is worth real thought — it is where you fix, permanently, the small frictions you'd otherwise correct in every conversation. That last sentence earns its place too: assistants lean agreeable by default (lesson 12 has more to say about this), and a standing invitation to push back measurably improves the pushback.

Memory that accumulates

Separate from instructions you write, most assistants now offer memory — facts the assistant saves from your conversations and reuses later, so mentioning the Nashik warehouse in March means it knows about Nashik in July. Convenient, and worth using with two habits. Review it occasionally — there is a settings page listing what's been remembered; facts go stale (the warehouse opens, the vendor gets replaced) and a stale memory produces confidently outdated answers, the worst kind. Curate it deliberately — you can usually tell it directly to remember or forget something. Treat memory as a notice board you own, not a diary it keeps.

Projects: a briefed room per topic

The most useful structure for work is the project (ChatGPT and Claude both use this name): a folder of conversations that share standing instructions and attached files. Priya builds one called Vendor management — instructions describing her vendor strategy and escalation thresholds, files: the rate cards and the contract summary. Every conversation inside starts already knowing all of it. A second one — Leadership reporting — holds her report template and last quarter's review, so "draft this week's report" is now genuinely a one-line ask; the five-part structure from lesson 3 didn't disappear, it moved into the walls of the room. Two or three projects around your recurring work is the sweet spot; the instinct to make twelve is the same instinct that makes twelve email folders, and it ends the same way.

One boundary to keep crisp: everything here — instructions, memory, project files — is standing context, which makes it standing shared data. Nothing goes into it that couldn't appear in a chat, which is the subject of the next lesson, and it's a short walk away.

Do this today: write your two-part standing brief and set it. Then create one project for your most repetitive stream of work and attach the two files you find yourself re-explaining most.