Pitch Ponies #5: Attack of the Hallucinating Sales Bot

A Supramono comic series about startup life, AI agents, and the chaos that follows when you skip the guardrails.

What you're about to read is a comic book story told in 7 slides, 20+ frames. Each frame below is written as a fully detailed image-generation prompt, so the visual sequence can be produced as a series of related illustrations. Read it straight through for the story, or hand the prompts to your image generation tool of choice.

This is the story of Unicorn AI — a living, breathing startup unicorn who demands to be fed money at all times. Unicorn is hungry. Unicorn is always hungry. And when CEO Mark decided to feed the beast by deploying an AI sales bot named Robot without guardrails, things went sideways in exactly the way every founder who's skipped human oversight has quietly feared.

AI hallucination rates vary widely depending on model, domain, deployment context, and how "hallucination" is measured — but the research consensus is clear that the problem is real, persistent, and consequential in live customer-facing deployments. Somewhere in that reality is where Unicorn's pipeline lives now. This is that story.

The real story behind the comic

Disclosure: This is a promotional piece produced by Supramono.

This is fiction. But the problem it illustrates is very real.

When AI generates product descriptions, specifications, or comparison data, fabricated details can create customer expectations that function like commitments — and in some jurisdictions and contexts may carry legal weight — resulting in broken trust when the product does not match what was described. That's not a hypothetical. Two documented cases illustrate the real-world consequences.

In Moffatt v. Air Canada (British Columbia Civil Resolution Tribunal, 2024), Air Canada's chatbot misrepresented the airline's bereavement fare policy to a customer, leading the tribunal to hold the airline liable for the misrepresentation — resulting in a judgment of approximately CAD $650. In Mata v. Avianca (S.D.N.Y., 2023), attorneys were sanctioned after submitting a legal brief that cited fictitious case law generated by an AI tool.

The scariest part of AI hallucinations isn't that they're wrong. It's how they're wrong. Research on AI calibration — including work such as Xiong et al. (2024) on the calibration of large language models — suggests that models can express high confidence even when generating incorrect outputs, using assertive language where hedging would be more appropriate. The phenomenon is well-documented in the literature, even if precise figures vary by model and context. Your AI sales bot doesn't hedge. It commits.

Multiple surveys of organisations deploying AI have found meaningful rates of reported negative consequences from AI inaccuracy — though exact figures vary depending on the survey population, methodology, and how "negative consequence" is defined. According to Zendesk's self-reported CX Trends survey data, a large majority of customer service leaders consider unresolved issues a significant driver of customer churn; readers should consult the primary report directly for exact figures, survey methodology, and population details. Robot closed 147 deals in a week. Mark lost goodwill in 50 calls — representing the most urgent cases, not the full customer count.

The fix isn't to avoid AI agents. The fix is to build them with guardrails, human-in-the-loop review for binding commitments, and verified product data that agents must stay inside. Experts and practitioners working on responsible AI deployment broadly recommend building verification into workflows from the start, rather than treating it as an afterthought — though the specific implementation will differ by organisation, use case, and risk appetite.

Unicorn is still hungry. It always will be. Just make sure what you're feeding it is real.

Pitch Ponies is a Supramono comic series about startup life, AI agents, and the chaos that follows when you skip the readme. New episode every two weeks.

Ready to build your AI venture engine — with guardrails included? Start free at Supramono and see how discover, build, and sell work when they're actually connected.