For Prompt engineer · 8 Grok Bots

A prompt bench that turns fuzzy requests into specs, evals, and traceable versions

8 Grok Bots for a prompt engineer: capture each request as a spec with acceptance cases, build a labeled eval set, draft prompt variants as diffs, record every run, and keep the version changelog traceable — all as drafts you review.

Hire the team See the bots
● Drafts only — nothing sends without your approval · Uses only your own data
After checkout
Your setup link arrives by email
We set up your fleet
After checkout, Workplace AI provisions your fleet on our Cursor Teams workspace and follows up soon. Watch for your Cursor invite and Grok Bot download link — then accept, install, and go.
8 bots·3 domain specialists·Success metric: Regression-caught-before-ship rate and prompt-version traceability·Category: AI roles
The problem it targets

Turning fuzzy behavior requests into shippable prompts — rewriting the same prompt from memory instead of diffing a version, hand-collecting test cases from transcripts every time, eyeballing outputs to guess whether a change helped, and losing track of which prompt version is live and why — so regressions ship unseen, "it worked yesterday" is unprovable, and every prompt change starts from scratch.

What you get

From day one

1

Every behavior request and prompt bug captured into a written spec with acceptance cases, and a labeled eval set built from your exported transcripts, so a change is measured against real cases, not memory.

2

Prompt variants drafted as a readable diff against the current version, and each exported run recorded per version so a regression shows up against the last version before anything ships.

3

Failing outputs clustered into root-cause notes, and a versioned changelog and weekly version report that ties every live prompt to its spec, its diff, and its run.

Meet the team

8 bots, each with one job and clear approval points

Read-only bots watch and prepare; drafting bots write but never send. You approve everything that leaves.

1

Chief of Staff

coach
Drafts for your review — you approve before anything leaves
Your morning brief, meeting prep, fleet roundup, and after-meeting coaching.
Approves: none — nothing leaves without you
2

Prompt Request Intake

intake
read-only
Watches & prepares — never sends
Capturing each request and bug into a spec with acceptance cases
Approves: duplicate/urgent flags for you to confirm
3

Eval Set Builder

researcher
read-only
Watches & prepares — never sends
A labeled eval set with a rubric, built from your transcript exports
Approves: none (nothing leaves; the set is internal)
4

Prompt Variant Drafter

drafter
specialist
Drafts for your review — you approve before anything leaves
Prompt variants as a diff against the live version, cases named
Approves: every variant before it is promoted
5

Regression Run Tracker

tracker
read-only
Watches & prepares — never sends
A per-version results table from your exported run logs
Approves: none (nothing leaves; you promote)
6

Failure Cluster Reviewer

researcher
read-only
Watches & prepares — never sends
Root-cause notes clustering the failing cases from a run
Approves: none (nothing leaves; internal analysis)
7

Prompt Library Reporter

reporter
read-only
Watches & prepares — never sends
The versioned changelog and a weekly version report
Approves: none (nothing leaves; you decide what ships)
8

Social & Referrals

growth
Drafts for your review — you approve before anything leaves
Draft posts and outreach to your own audience, credited and approved by you.
Approves: sends
The handoff — one shared chat
Intake→Builder→Drafter→Tracker→Reviewer→Reporter
What the fleet does in week one

All read-and-prepare — you check its work before anything is trusted

What it does not do

The guardrails are the product

  • ×It drafts and records; it does not ship. Every prompt variant and version note waits for your promotion.
  • ×It never runs an eval against a live model endpoint or spends on inference — it reads the run logs you export.
  • ×It never fabricates an eval result, a pass rate, or a case label, and it never promotes a prompt to a live setting.
  • ×It reaches only your own opted-in audience with version notes you approve, and never publishes on its own.
Regression-caught-before-ship
rate and prompt-version traceability
How you'll know it's working

One number, measured over time

Share of prompt changes that carry a labeled eval result before promotion, and how many live prompt versions can be traced to a spec, a diff, and a run in the changelog

Before day one

What you fill in — the bots never guess these

eval-set
the header contract for the labeled eval set every bot reads: input, expected behavior, rubric line, source transcript.
run-results
the per-version results table the tracker fills from your exported run logs. Scores come only from here.
prompt-variant
the layout for a diffable prompt variant with the acceptance cases each edit targets.
Pricing

Cancel anytime.

Workplace AI provisions and runs your CrokPot fleet on our Cursor Teams workspace. We invite you on your own email — you accept, install Grok Bot, and we handle the rest.
$200/mo
  • Your whole fleet provisioned and installed for you
  • Cursor Teams seat included — paid by Workplace AI, never billed by Cursor
  • Onboarding + ongoing training
Hire the team
Your Cursor Teams seat is included and paid by Workplace AI. Drafts only, using your own data. No earnings, time-saved, or outcome guarantees are made.
How access works

You buy a CrokPot fleet from Workplace AI — not a Cursor plan

CrokPot runs on Workplace AI, Inc.’s (WAI) Cursor Teams workspace. WAI keeps the unpaid admin seat and pays Cursor for your seat. You join as the invited user on your real email and work through the installed Grok Bot fleet.

Your card is charged by WAI via Stripe for the CrokPot fleet ($200/mo, auto-renewing monthly) only — never by Cursor. If you cancel your fleet subscription, the invite/seat ends with it and is not portable to your own Cursor account.

FAQ

What you get, who owns the seat, and what happens after you pay

From the team behind Ancestry — CrokPot is built by Paul Allen, a co-founder of Ancestry.com.

Drafts only — you approve everythingYour own data, never shared$200/mo · cancel anytimeProvisioned & supported by Workplace AI

What do I actually get?

A pre-built fleet of Grok Bots for your trade, provisioned and installed for you on Workplace AI’s Cursor Teams workspace, plus onboarding and ongoing training. Every bot drafts; you approve.

Who owns the seat?

Workplace AI, Inc. (WAI) owns the Cursor Teams account and pays Cursor for your seat. You’re invited as a paid user on your own email. The seat isn’t portable to a personal Cursor account.

Do the bots act on their own?

No. The fleet is drafts-only, working from your own data. Nothing sends, files, or is trusted until you review and approve it — that’s the whole design.

Can I cancel?

Yes, anytime. It’s an auto-renewing monthly subscription at $200/mo. If you cancel, your invite/seat ends with the subscription.

What happens after I pay?

We follow up soon: our team provisions your fleet and invites you on the email you used at checkout. Watch for your Cursor Teams invite and Grok Bot download link, then accept, install, and go.

Training

New to Grok Bots? Get your team trained.

Live sessions and replays to get you and your coworkers productive with your fleet fast.

Explore Grok Bot training