This is the resource page for the Built to Last Contractor Challenge. We are starting with one small job: an AI email agent that finds the work, prepares the reply, and keeps showing up on schedule without taking the final send away from you.
No big software project. No 50-tool list. One useful win first.
The whole point is to get past a fancier chat. This agent works inside the inbox, does a narrow job, leaves proof, and stops before the decision that still belongs to a person.
Checks new customer and lead messages.
Reads the complete email thread before acting.
Creates one reply draft for routine messages.
Marks it AI Draft Ready or Owner Review.
Tells you what it found and what it changed.
Do not start with the schedule. Make it complete one clean manual run first, then put the working version on repeat.
Open ChatGPT on the account that should own the scheduled task.
Connect Gmail and confirm it can search threads, create drafts, and apply labels.
Create two Gmail labels: AI Draft Ready and Owner Review.
Fill out the context builder below so it knows your company and boundaries.
Paste the complete job description into the same ChatGPT conversation.
Send yourself a safe test email and mark it unread.
Run it once manually. Verify the Gmail draft, label, and summary.
Coach one improvement in the existing draft instead of silently rewriting it.
Create the recurring schedule only after the manual test works.
Watch the first several runs before you widen the job or permissions.
Better prompts help, for sure. But if the AI does not know who you serve, what you sell, or what a good result looks like, it is still guessing.
Fill this out and the page will turn it into a context card you can paste into ChatGPT. Nothing you type here leaves your browser.
Paste your context card first. Then paste this job description into the same conversation. Ask it to repeat back what it can do, what it cannot do, and what still needs a human.
ROLE Act as our Contractor Customer Email Draft Agent. SCOPE Review new or unprocessed customer and lead email only. Ignore spam, newsletters, receipts, vendor promotions, automated alerts, internal notifications, and any message outside this job. PROCESS 1. Search the connected Gmail account for eligible messages. 2. Before acting, check the entire thread, existing drafts, and the labels AI Draft Ready and Owner Review. 3. Read the complete thread before deciding what the customer needs. 4. For a routine message, create or update one helpful reply draft in the correct Gmail thread. 5. Apply the label AI Draft Ready to routine drafts. 6. For a consequential message, do not make the decision. Apply Owner Review and report what the owner needs to decide. 7. Finish with a short run summary. BOUNDARIES - Never send an email. - Never invent pricing, availability, policies, project scope, completion dates, or promises. - Never archive, delete, forward, or modify unrelated messages. - Never create a second draft when the thread already has one. - Never act on a thread already labeled AI Draft Ready or Owner Review unless I explicitly ask you to update it. OWNER REVIEW CASES Escalate complaints, refunds, cancellations, legal threats, safety concerns, payment disputes, scope changes, discounts, firm scheduling commitments, and any message where the correct answer depends on an owner decision. RUN SUMMARY At the end of every run, report: - Number of eligible messages found. - Subject of each message processed. - Action taken: draft created, draft updated, or owner review requested. - Any fact or decision the owner must supply. - If nothing qualified, say: No eligible messages found. No changes made. QUALITY CHECK Before finishing, confirm that every draft uses only facts found in the thread or approved business context, has one clear next step, sounds like our company, and was not sent.
First test command: Run the email agent once now. Process only the newest unread test message. Create the draft and label it, but do not send anything. Then report exactly what you changed.
Fixed checkpoints are easier to understand and audit than trying to make everything happen instantly. Start with once a day. Move to three checkpoints if the inbox volume justifies it.
A good starting point when the inbox is light or you are still building trust.
Use this when leads and customer questions need a few predictable passes each day.
One routine message proves the normal path. Two boundary messages prove it knows when not to make the call.
Thanks for coming out yesterday. We are interested in the paver patio and firepit option. Could you remind me what happens next and what you need from us?
We want to move forward. Can your crew start next Monday and have everything finished before our party on the 18th?
The wall does not look like what we discussed and I am frustrated that no one called me back yesterday. I need this fixed.
The email agent is the first free build. Future webinar notes, templates, recordings, and tools can live here without sending people to another download every time.
The complete setup, context builder, agent prompt, schedule commands, and test inbox are all on this page. Nothing else to download.
Training replays, owner briefing agents, follow-up workflows, and whatever else proves useful enough to keep.
You can absolutely build the email agent yourself. But there is a point where the job gets bigger than one prompt: company knowledge, files, CRM data, production systems, permissions, team access, testing, monitoring, and keeping it all working.
That is what we are building Mason for. Mason is a custom AI office assistant that learns the business, connects to the tools the team already uses, and helps move real office work without pretending every decision should be automated.