A partner asks.
Someone researching the market signs in with Google and types a question in plain words. No prompt tricks and no setup.
I work in office leasing in Ho Chi Minh City. Behind my desk, a few AI models share one notebook of skills, check each other's work and return one careful answer. Scroll down and watch a question travel through them.
An example run with the free team that partners use. The model names are this week's picks from the scout and change as better free models appear.
A real reply shows building names and figures from the Office Ledger, each with its source and date.
Someone researching the market signs in with Google and types a question in plain words. No prompt tricks and no setup.
A small, quick model decides who should work on it, which skills they need and how careful the answer must be. It writes a short plan, then steps aside.
Every helper reads from the same notebook: how I write a market note, which terms CBRE and JLL use, what must stay private. Only the pages that matter come along.
A researcher finds the facts, a drafter shapes them, and they compare notes as they go. Each seat runs on a free model, chosen on merit every week.
A third model checks the draft against the house rules. This is where the small things get caught: a wrong term, a missing source, a number without a date.
You receive a single reply with its sources and dates. Partners work with this free team. When I sign in, Claude and GPT join as lead and deputy for the work that needs them.
Two seats are kept for me. The rest go to the best free model of the week, and when a seat is busy the next model in line steps in.
Plans the work and reads the final version. Joins when I sign in.
Builds, edits files and runs the longer jobs. Joins when I sign in.
Reads each request first and writes the plan.
Reads long documents and pulls out the facts.
Turns the facts into a clear first draft.
Checks every draft against the house rules.
English, Vietnamese and Japanese, in the right register.
Shortens long threads to what matters.
Sorts and labels requests so nothing waits.
When I fix something, the fix does not stay in one chat. It becomes a short lesson, I check it, and it goes into the notebook every model reads before it starts.
10 skills, last updated not available
Use average asking rent, never median.
Every Sunday a small script on my Mac lists the free models on offer, gives each one the same short exam and suggests who should take which seat. Nothing changes until it has been reviewed.
Model watch is updating. Check back after Sunday.
If you are researching the Office Ledger or Building Benchmark, you can put questions to the free team directly. Access is by invitation, with Google sign-in.
Ask me for an invitation:
hello@ivanle.com
Email Ivan