Back to the Goal Mode framework
She was selling 35 dahlia tubers a week at her best. That was the baseline. That's what years of running her farm, posting on social media, and managing her own email list had produced.
We gave Codex a goal, a deadline, and 30 minutes of setup. Thirty-six hours later, she had sold 490. Over $4,000 in revenue. Her previous best week, multiplied more than 10 times by a system she walked away from after half an hour.
The question isn't whether Codex can sell. The question is whether you're willing to give it the access and the deadline it needs to prove it.
This is the case study I walk every business owner through when they're deciding whether goal mode is worth attempting. Not because the numbers are impressive, though they are, but because the setup is completely replicable. There was no special technology. No custom integration. No developer. Just a well-scoped goal, the right access, and the discipline to walk away.
Watch Heather's story video if you want to see the full breakdown in context.
Who Heather Is, and Why We Started With Tubers
Heather runs a B2B people operations consulting firm for growing manufacturing companies. That is her primary business. She also runs a dahlia farm as a side business, 14 acres, seasonal inventory, and a customer base she'd built over years.
We did not start with her consulting firm. We started with the tubers.
That sequencing matters. In a B2B consulting practice, a single mishandled outreach email can cost a relationship worth tens of thousands of dollars. In a dahlia farm side business, a poorly timed Facebook post costs nothing. The risk profile is completely different. When you're calibrating an agentic system for the first time, you calibrate it where the cost of error is lowest.
If you're reading this and thinking "but my main business is the one I care about", that's exactly why you start somewhere else. The first few goal runs are for learning, not for closing. Read more about the low-risk sequencing framework.
How the Goal Was Structured
This is the part most business owners skip over or gloss past: the specific configuration choices that determined everything that followed.
| Component | What Was Set | Why It Mattered |
|---|---|---|
| Target | Sell 490 tubers | Hard but achievable given actual inventory, not the full supply |
| Deadline | Sunday at 11:59 PM PT (~35 hours) | Created urgency; forced Codex to act rather than plan indefinitely |
| Access granted | Shopify, Gmail (main + second account), Facebook via browser | Every tool a human salesperson would need to close this sale |
| Parameters | Send a few emails per day; engage on Facebook; create personalized time-limited coupons | Defined the fence without micromanaging the route |
| Setup time | Approximately 30 minutes | Total human investment before walkaway |
Notice what is not in that parameters column: which Facebook posts to write, what the subject lines should say, which customers to contact first. Codex made those decisions. The parameters set the boundary conditions. Execution lived inside them.
The target was set at 200, not 500, not the total inventory. Codex needs a number that is genuinely difficult but reachable given what's available. A number that's mathematically impossible to hit defeats the system before it starts. A number that's trivially easy removes the urgency that drives agentic behavior.
The deadline was approximately 35 hours. Short enough to create pressure. Long enough to actually work across email, social, and direct contact cycles.
What Codex Did, Without Being Told How
Once the goal was active, Codex left a running action log. Here is what it did, unprompted and without individual output approval:
- Posted to Facebook multiple times throughout the day
- Found Dahlia-specific Facebook groups that were outside the original scope and requested permission to post there, Heather granted it
- Contacted past customers with personalized, time-limited coupon codes
- Identified a phishing scam email impersonating Shopify and flagged it
That last one is worth pausing on. Codex wasn't asked to monitor for security threats. It encountered an email during the course of operating in Gmail and identified it as a phishing attempt. That kind of contextual judgment, showing up inside a focused sales goal, is what separates an agentic system from a scheduled automation.
The Facebook group expansion is equally instructive. Codex found communities its access didn't originally include, assessed them as relevant, and asked for permission before acting. One decision required from Heather. One. In a 35-hour, $4,000 goal run.
The Result, and What It Tells You
490 tubers sold by the deadline. Goal exceeded. Codex stopped on its own, displayed a congratulatory message, and waited for a new instruction.
Her best week before this was 35 units.
The math is not subtle. But the lesson isn't in the math, it's in the mechanism.
This wasn't a marketing breakthrough. Heather didn't discover a new channel or invent a new offer. She gave an agent the access it needed, a hard target, a real deadline, and the space to work. The 490 tubers are the direct output of that configuration.
The common mistake I see business owners make: they look at a result like this and think "I need better AI." What they actually need is better setup. The AI Heather used is the same AI available to everyone. The difference was in how the goal was structured.
Why the Side Business First Principle Protects You
Running the first goal on a low-stakes asset isn't timid. It's strategic.
Your first few Codex goal runs are calibration runs. You are learning what parameters Codex needs. You are learning how it handles edge cases. You are learning what it does when it hits a boundary you didn't anticipate. You are learning where your permissions setup has gaps.
None of that learning should happen on your most important relationships.
Heather's dahlia farm gave her a clean, low-risk environment to see exactly how Codex behaves when operating autonomously. Now she has a documented playbook. A successful goal run, with parameters she knows work, that she can adapt and redeploy for any future event, Father's Day, fall planting season, whatever the occasion calls for.
That playbook is now an asset. The Mother's Day run doesn't get archived. It gets refined and reused.
For the complete framework on how to structure goals and sequence your first runs safely, read the full guide.
The One Setup Decision That Made This Possible
Thirty minutes. That's what the access configuration took before the goal was activated: connecting Shopify with coupon creation permissions, linking two Gmail accounts, and logging into Facebook via an open Chrome browser.
Most business owners underestimate this step. They either skip parts of it, rush through it, or assume Codex will figure out the access it needs mid-run. It won't. What Codex can access at setup is the ceiling for what it can do during execution.
Think of it this way: if you hired a salesperson and didn't give them a phone, a customer list, or the ability to send emails, you wouldn't blame them for not closing deals. The tools have to be there before the work can happen.
Heather's 30 minutes of setup gave Codex everything a human salesperson would need to run that campaign. That's the standard to aim for: ask yourself what a competent human would need to complete this task, then make sure Codex has access to every one of those things before you activate the goal.
If a plugin reports as unavailable after setup, that is a permissions issue to resolve before running a goal, not a problem to push through. Learn how to troubleshoot Day 1 permissions errors.
What to Do Before You Set Your First Goal
The case study is useful. The replication is what matters.
Before you activate your first goal, work through this checklist:
- Identify a low-stakes asset, a side business, a lower-ticket product, a non-primary audience
- Write the goal in one sentence: specific number, specific deadline, specific outcome
- List every tool a human would need to complete this task, that's your access list
- Connect each tool before activating, Shopify, email accounts, social platforms
- Set a 24-48 hour deadline, no longer for the first run
- Ask Codex to show you the execution plan before it begins, approve it, then walk away
- Read the action log after the run ends, not during
The baseline you get from that first run, even if the goal isn't fully reached, is more valuable than any plan you could write in advance. You'll see exactly what Codex does with the parameters you gave it. That data shapes everything that follows.
Heather's tubers aren't the story. The 30 minutes of setup that made them possible is. That's where the result was built., Shanee
What the Case Study Actually Proves
The number people remember is 490. The number that matters operationally is 30: the minutes of configuration that made the run possible. The result did not prove that every business should hand its best relationships to an agent on the first weekend. It proved the opposite. Start where mistakes are cheap, document the action log, then carry the playbook into higher-leverage parts of the business once you have evidence.
The hard stop matters too. When Codex reached the 490-tuber result, it stopped and waited for Heather to set the next goal. Goal Mode is autonomous inside the configured goal. It is not supposed to invent the next business objective on its own.