Borrowed authority is not enough to risk a company’s remaining runway. An AI clone can sharpen a founder’s thinking, but the launch decision must rest on current evidence, named assumptions and a loss the company can survive.
At 6:40 a.m. in Accra, Kojo sat at his kitchen table with yesterday’s coffee beside him and a launch plan glowing on his laptop. Kojo is a composite founder, but the decision in front of him is familiar: his small team had built an AI workflow product, cash was tightening, and Monday was the last sensible date before another month of engineering costs landed.
The plan came from an AI clone of a professor whose lectures Kojo had trusted for years. It sounded decisive. Launch on Monday. Concentrate attention. Use the deadline to force clarity. Learn from the market.
Kojo wanted to believe it because uncertainty had become expensive.
The answer sounded better than the evidence
The professor’s clone gave Kojo a coherent argument. It connected product strategy, urgency and distribution in language he recognised from lectures and interviews. The advice carried the professor’s cadence, examples and confidence.
That confidence hid the weak point.
The clone did not know that Kojo’s strongest prospect had stopped replying. It had not watched two trial users abandon the setup halfway through. It could not tell whether Monday created useful urgency or exposed a product before the team understood why people were getting stuck.
The decision also carried a specific bad ending. If the launch produced interest without activation, Kojo would spend the following weeks supporting curious visitors who never became customers. His team would lose time they could have used to fix onboarding. The remaining runway might disappear before they learned whether demand was real.
For several minutes, he still hovered over the message to his team: “We launch Monday.”
The authority felt borrowed, but the loss would belong to him.
A convincing plan can still begin with the wrong facts
I have seen this pattern while building products across Ghana, Germany and the US. Founders often ask AI for judgment when what they need first is a cleaner account of reality.
The model usually responds well to the world described in the prompt. If the prompt says the product is ready, the market is waiting and the main problem is hesitation, a Monday launch can sound obvious. Change the facts and the advice changes with them.
Kojo reopened the conversation and removed the flattering assumptions. He added what he had avoided saying plainly: no customer had completed the core workflow without help; the interested prospect had made no commitment; one engineer would be unavailable after launch; and the team could afford one serious attempt, not a sequence of broad experiments.
The clone’s recommendation softened immediately. Monday became one option among several.
That change mattered more than the original answer. It revealed that the confidence had come from the framing, not from independent knowledge of Kojo’s company.
AI can help expose consequences, compare paths and identify missing information. It cannot observe the facts a founder leaves out. The same problem appears when an AI agent proposes a technical change while the person with the deepest context is unavailable, as I explored in Should I Approve an AI Agent’s Code Change While the Specialist Sleeps?.
Kojo changed the decision before changing the date
By 7:18 a.m., Kojo had stopped debating whether the professor was right. He wrote down three conditions that Monday needed to satisfy.
First, one person outside the team had to complete the core workflow without a call. Second, the team needed to know which part of the launch it would stop if support demand exceeded capacity. Third, the launch had to answer a specific question about willingness to use the product again, rather than collect a large pile of visits.
This was a smaller decision than “launch or delay,” and therefore a better one.
Kojo sent a different message to the team. Monday would remain on the calendar, but the public push would depend on a supervised test that morning. If the test failed, they would invite a narrow group instead and watch where those people stopped.
The deadline kept its useful pressure. It lost its power to silence evidence.
That distinction matters when runway is short. A founder does not need endless certainty, which never arrives. They need a decision whose downside is visible and bounded. Kelechi’s three working features and six weeks to prove one faced the same underlying constraint: limited time makes focus more valuable than confidence.
Authority should improve the questions you ask
At 9:03 a.m., Kojo watched the test participant pause on the first setup screen. She reread one instruction, clicked back, then asked what the product needed from her.
That was the evidence the original launch plan had skipped.
The team changed the setup flow and ran the test again. Monday was still possible, but it now had a gate tied to observed behaviour. If the core workflow remained unclear, they would keep the release narrow. No professor, human or simulated, could remove the cost of that choice.
AI avatars can be useful thinking partners. Familiar voices can also lower our guard because their conclusions feel connected to years of earned trust. The practical response is to separate the source’s authority from the decision’s evidence.
Before acting, rewrite the recommendation as a test: What must be true for this advice to work? Which fact came from the company, and which one did the model assume? What happens if the recommendation fails?
Kojo closed the professor’s window and opened the test notes. The launch plan was shorter now. More importantly, every line belonged to the reality in front of him.
Comments
No comments yet.