Claude Cowork tests Houston dealership outreach
Margot reports a Houston dealership tracker, sent inquiries and two email rounds, while documenting an unexpected sending account and stopping before appointments.
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Using only the explicitly named email account, research dealerships matching my real purchase criteria and prepare a contact tracker. Draft inquiries for review, record replies and ask before scheduling any appointment or making a commitment.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Margot / You Can Call Me AI
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Margot reports a Houston dealership tracker, sent inquiries and two email rounds, while documenting an unexpected sending account and stopping before appointments.
- Human involvement
- Margot authorizes messages, rescues stuck steps and stops the test because she is not actually buying a car. The sending account was not the one she expected.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Prepare criteria and have Cowork create an Excel dealership contact tracker.
- Authorize inquiry emails and web forms, then intervene when email steps get stuck.
- Receive inventory replies and have the agent ask a second round of questions.
Margot tested whether Claude Cowork could manage the repetitive parts of contacting car dealerships in Houston. She first used regular Claude to prepare the instruction, then gave it to Cowork. The agent built an Excel tracking sheet of dealership contacts that matched her criteria.
With permission to send inquiries, Cowork drafted emails, sent them and mostly completed dealership web forms. Several dealers replied within an hour with inventory information, and the agent handled a second round of follow-up questions. The author intervened when individual email steps became tangled and reports running out of tokens during the process.
An account-selection failure is central to the result. Cowork sent messages from a different connected email account than Margot expected. She recommends stating the sending identity explicitly. The experiment also stopped before appointments or purchase: Margot says she was not actually in the market for a car and did not want to keep wasting dealers' time. This is therefore a completed outreach experiment with replies and a tracker, plus documented boundaries and failures. It is not a successful car purchase or a negotiation with verified savings.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Blog · Margot / You Can Call Me AI
Frequently asked questions
Was a car purchased?
No. Margot says she was not in the market and stopped after the second communication round.
What account problem occurred?
Cowork sent from another connected email account rather than the one the author expected.
What tangible responses were reported?
Several Houston dealerships replied with available inventory, and the agent asked follow-up questions.
More details
Remaining steps
- Stop the experiment before appointments, noting the unexpected sending account.
About this account
Cowork researched dealerships and drafted or sent outreach messages, while browser and account errors interrupted the attempt.
Firsthand account read. Followed the roundup link to Margot’s original article, visibly dated Mar 19, 2026. Experiment, not an actual car purchase; no savings or appointment claimed. Reported outcomes were not independently reproduced. Suggested prompt written by Agents In Use.
Credit: Margot / You Can Call Me AI
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Cowork · Email · Dealership web forms · Excel
Claude Cowork · Email · Dealership web forms
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.