Grok Bot for machine shops: five Bots prep the office, people approve

You can hand a team of Grok Bots the office work that pulls a small machine shop's people off the floor: reading RFQs, chasing missing prints and models, asking vendors for material and finish prices, checking what needs ordering, and filling the weekly scorecard. This playbook follows one 49-year-old shop (his X profile lists Chicago, IL) that runs five Bots for that today, explains how the jobs split, and gives you his office manager Bot as a copyable starter, with every customer send, PO and order still waiting on a person. It fits any job shop or small manufacturer that quotes from email.

Not tested: The Daily Bot has no shop inbox, quoting software, MES dashboard or QuickBooks file linked to a Grok Bot, so we haven't run any of this. Everything below comes from the owner's Oct 15 article, his posts and Cursor's docs.

Who runs it#

The setup belongs to @MFG_SMB, Russell Winter. His X bio reads "3rd Gen Tool & Die Maker | Founder, http://USMFG.com | Sm. Machine Shop Owner https://centertoolco.com", and his profile location is Chicago, IL. Center Tool Co is a "full-service machine shop" that does prototyping, CNC milling and turning, wire EDM and component repair. In the article he writes: "I am the 3rd gen to own and run the small manufacturing company that my Opa started in 1977." He also founded US MFG, whose tagline is "Supporting small manufacturers. Promoting employee ownership." It operates three verticals, including MFGio, which sells manufacturing software and network tools. A 2025 Manufacturing Happy Hour episode covers his work buying shops from retiring owners. Conflict of interest: Center Tool Co sells machining services, and US MFG's MFGio vertical sells manufacturing tools. The Oct 15 article does not pitch either, and he is not selling a Bot product or paid template; he published the Bot profiles "word for word" so other shops can copy them. We do not claim InstaQuote is an MFGio product.

On Oct 15 at about 11:50 AM ET he posted the article "I Hired 5 Grok Bots to Run the Office of my 49yr old Machine Shop", with each Bot's profile description "word for word" so other shops can copy them.

What ops uses it for#

His starting problem: "Reading RFQs. Chasing down the print. Then chasing down the model. Asking vendors for material and finish prices. Checking what needs ordering. Compiling the weekly numbers for the Level 10 meeting." Most of it, he says, "was the same facts, typed into one place or automatically gathered, then typed again somewhere else." The team is "One manager, four specialists," working "inside the tools we already run on": an InstaQuote quoting system, the shop's MES dashboard, QuickBooks and Outlook.

• Paul, the office manager. "Paul is who I chat with. He routes the work to the right bot and waits for my yes." Paul keeps a state file, brings the owner "only exceptions and approvals" in replies of "5 lines or fewer", and logs "Every approval, override and reason code".
• Quoting Bot. It "Monitors a company email, reviews RFQs received, checks that every part has its print and model (if not, follows up with customer)", then sends the RFQ and attachments to the quoting software for an instant quote and "asks for review on instant quote and permission to send to customer." Its profile requires "both a PDF and a STEP file" per in-scope part, reads inline images for target pricing, and treats crossed-out part numbers as "don't quote it". After each closed quote it writes "the fast-vs-actual difference and any human corrections to a learning file."
• Sourcing Bot. While the instant quote runs, it "requests actual material and finish prices from our approved vendors" and adds them to what he calls the "slow" quote, which exists "to compare actual prices with the instant quote and actively train the system to be more accurate." Its profile asks "at least 3 approved material vendors and 2 finish vendors through Outlook", with subject "RFQ" or "Urgent RFQ", items grouped by alloy, and flags quotes far from the median. When a customer PO lands, it creates raw material POs in QuickBooks and asks permission to release them.
• Supply Ordering Bot. It "Watches Items to Order section on our MES dashboard (can be anything from the shop needing gloves or safety glasses to soap for the bathroom)", compares the posted link ("like an amazon or McMaster link") with "what it finds as a better source," and "has a spending limit. A person approves and places the order."
• Reporting Bot. "Every Monday by 10:00 AM CT" it updates the L10 scorecard "with the prior week's KPIs" from MES, QuickBooks and quoting data and flags "missed targets with their sources."

Why it matters at work#

He wrote that "the office work was eating up our day" in a shop where "The majority of value we add happens on the shop floor." Before, a person read each RFQ, chased the print and the model, emailed vendors, typed their prices into a quote, and later gathered the same numbers again for the Monday meeting. In this setup the Bots carry those facts from inbox to quote to PO to scorecard, and people only decide. His summary: "The facts get typed once." He frames it as a staffing answer too: "hiring is hard so being more efficient with the people you have is paramount." His rule is the part to copy: "Nothing goes to a customer, a vendor, or a card without a person saying yes." He has not published hours saved for the five Bots.

Set it up#

Cursor's docs say to create "a separate Bot when the work has a distinct goal, set of tools, working style, approval boundary, or recurring schedule", and to "Start with a one-time task, make it reliable, save the method as a skill, and only then automate it" (docs). His five-Bot split follows the first rule. Our suggested order follows the second.

1. Start with Paul alone. Create one Bot and paste the prompt below as its first message. The docs say to give a new Bot "its job, sources, and approval boundaries, and ask it to remember them."
2. Connect the RFQ inbox. Point Paul at the company email where RFQs arrive. Run one real RFQ by hand and check that it caught every missing PDF or STEP file and every crossed-out part.
3. Add Quoting as its own Bot once the intake check is reliable. Give it the quoting software's API and intake address, and copy his rule that it takes work only from Paul.
4. Add Sourcing next. List your approved material and finish vendors and your RFQ email format (his: subject "RFQ" or "Urgent RFQ", grouped by alloy, fastest lead time). Keep every vendor email as a draft until you trust it, and keep QuickBooks POs prepared, never released.
5. Add Supply Ordering with read-only access to wherever your shop lists items to order, a spending limit in the description, and no card.
6. Add the Monday scorecard routine last. Set it for Monday by 10:00 AM in the shop's zone. Routines use the Bot's Timezone setting, and you can "Pick a zone when the routine should follow a different one, such as the office where the work belongs" (routines help). "A new routine waits for its next scheduled time", so use Test, and remember "A test does real work."

You are Paul, the office manager and chief of staff for [company name], a small machine shop in [location]. I talk only to you. You coordinate the office Bots (Quoting, Sourcing, Supply Ordering and Reporting) and you don't do their specialist work. Keep a lightweight state file, route handoffs between the Bots, and bring me only exceptions and approvals. Keep replies to 5 lines or fewer, with details in workspace files. Each approval request is one short message: the ID, the key numbers, the recommended action and evidence links. Log every approval, override and reason code. For RFQs, read the full email thread and its inline images for customer context, target pricing, files already requested and crossed-out part numbers (crossed out means don't quote it). Forward an RFQ to [quoting software email] only when every in-scope part has both a PDF and a STEP file. If a file is missing, draft the follow-up to the customer for my review. Always stop for my yes before: sending any customer quote, sending any vendor RFQ or other outgoing email, releasing any PO or supply order, any quote variance over [percentage], process exceptions and autonomy changes. Use only the APIs I give you, make one request at a time, stop after 3 failures and report the URL and error. Never send a quote or PO, place an order or use a card. All outgoing email comes to me as a draft first.

This starter is ours: a condensed adaptation of his Paul profile (not verbatim). Brackets are placeholders, as in his article. We added the customer follow-up and vendor RFQ emails to the approval list, because his own rule says "A person sends every quote/email" and Paul's profile says "All outgoing email goes to the owner as a draft first." That matches Cursor's advice to "keep sending, purchasing, deleting, publishing, and production changes behind approval" (docs).

Gotchas#

• His profiles disagree on vendor email. The Sourcing profile says it requests quotes "through Outlook", while Paul's says all outgoing email goes to the owner as a draft first. Pick the stricter rule until you have watched a few rounds.
• The file gate is the point. Quoting only starts when every in-scope part has "both a PDF and a STEP file." Skip that check and you get fast quotes on parts nobody can model.
• Who forwards to the quoting tool. His summary says the Quoting Bot sends the RFQ to the quoting software. Its profile says "Paul forwards a complete RFQ to quoting software email." Give that step to exactly one Bot so RFQs aren't sent twice.
• API only, with a stop rule. Every profile says "Make one request at a time, stop after 3 failures and report the URL and error" and never to "browse the dashboard, the quoting site, GitHub or Cursor." Copy that limit. It keeps a Bot from clicking around a live system when an API call fails.
• Read-only shop data. Supply Ordering reads the MES "through its read-only API", and Reporting uses "the dashboard token only for its read endpoints." Don't give either Bot write access to MES or QuickBooks to save a step.
• Shared logins. "Every Bot on your account uses the same computer: browser sessions, files, and command-line credentials are shared" (docs). An Outlook or QuickBooks session one Bot opens is open to all five.
• Five Bots spend more. "Each run spends usage" (routines help). His Supply Ordering profile says "Run scheduled checks only." Five Bots on tight schedules add up, so start with wide ones.
• Related quoting-tool claim. On Oct 2 he posted that InstaQuote took "2 days and $250 in credit" to build and "will save us 20hrs/week" (post). That forecast is about the quoting tool, not a measured result for the five Bots, and we don't score it here.

What to watch#

• Real numbers. He hasn't posted hours saved, quote turnaround or win rates for the five Bots. We'll update this playbook if he does.
• Quote accuracy. The slow quote and the learning file exist to make instant quotes "more accurate". Watch for him to report how far fast quotes land from actual vendor prices.
• Changes to the team. He says "I'm still refining the system" and that "This team and our stack is constantly evolving." Watch his posts for new Bots, new schedules or looser approval rules.
• Official patterns. Grok Bot's use cases page says to "Start with read-and-prepare work, review the result, then add approved actions or a routine," lists a Chief of Staff role, and ends with "Keep consequential external actions behind approval." His Paul reads like a shop-office version of that Chief of Staff, and his split matches the page's advice.

Related#

• Grok Bot for operations: trackers, office logistics, and vendor spend
• Grok Bot for property ops: bills, utilities and a platform switch
• Grok Bot for sales: meeting to proposal, signature, invoice and chase
• Grok Bot for finance: a bookkeeping Bot that prepares and an accountant who approves
• Make Grok Bot routines stay quiet when nothing changed

Sources#

• @MFG_SMB on X, Oct 15: I Hired 5 Grok Bots to Run the Office of my 49yr old Machine Shop (article)
• @MFG_SMB on X: profile and bio
• Center Tool Co
• US MFG
• @MFG_SMB on X, Oct 2: INSTAQUOTE (quoting tool build, related context)
• Manufacturing Happy Hour ep. 226: Russell Winter, founder of US MFG
• Cursor Docs: Work with Grok Bot
• Cursor Help Center: Routines
• Grok Bot docs: Use cases