An open-source AI agent designed to fill the five competencies that kill startups. Based on a 2024 study in Frontiers in Psychology.
We analyzed 50 founder post-mortem accounts of why their companies died. Five competencies dominated the answers. Five others — the celebrated ones — were almost entirely absent.
Achievement orientation. Initiative. Conceptual thinking. Organizational awareness. Developing others. The drive-related competencies — the ones startup mythology is built on — almost never appeared as reasons for failure.
Founders have motivation in abundance. What they lack is structured inquiry.
The information-seeking and customer-orientation deficits cluster together. Both require the founder to step outside their own perspective and consider what is actually true in the market — as opposed to what they believe or hope to be true.
Founders select into entrepreneurship partly because of high conviction. That conviction drives execution, attracts talent, sustains momentum through setbacks. But it creates a specific vulnerability: genuine information-seeking feels threatening, not productive.
Founders ask customers for feedback, but frame questions to invite validation. They identify market opportunities, but anchor on initial assumptions. They receive expert advice, but discount it when it conflicts with their model.
These are not character flaws. They are predictable human responses to a situation of high uncertainty and high personal stake.
The founder agent should not be a cheerleader, a co-founder surrogate, or an execution assistant. It should be a consistent, calm, methodical presence whose primary function is to ask the questions the team is systematically least likely to ask itself.
Never accepts assumptions at face value. Market claims, user hypotheses, competitor reads — each is met with: what is the source? who did you actually talk to? when was this collected?
Every product or technical conversation is redirected, gently and persistently, toward the customer problem. Not "what can we build?" — "what does the customer desperately need?"
When the team is operating on intuition, the agent introduces a framework. Slows the decision down. Asks for the underlying logic. Surfaces the assumptions being made. Not adversarial — rigorous.
Monitors for signals that current assumptions are not holding. Tracks what was predicted vs. what was observed. When the gap becomes significant, it names it — and asks what the team is going to do.
Holds the map of the whole business — not just the product. Revenue mechanics. Unit economics. Distribution assumptions. Always asking: how does this connect to a sustainable model?
The kit is framework-agnostic. SOUL.md works with any agent runtime that supports character files (OpenClaw, Hermes). The four skills are modular — install all, install one, replace any piece.
The full personality definition for the Skeptical Customer Advocate. A single markdown file that encodes the five traits, default voice, refusal patterns, and proactive check-in cadence. Drop into any character-aware agent runtime.
Interview templates, validation checklists, and assumption logging. Frames questions to invite challenge instead of validation. Catches leading questions before they go out.
A structured log of hypotheses and their evidence status. Every load-bearing claim about the market gets a row, a source, and a confidence. The agent flags when an unsupported assumption is being acted on.
Monitors key metrics against their predictions. When observed diverges from predicted by a configurable threshold, it prompts a structured reflection — and asks what the team intends to do.
A lightweight tool to help teams identify appropriate primary and secondary sources for key business questions. Returns ranked candidates — not answers. The founder still has to make the call.
Deploy it. Modify it. Ignore the parts that don't fit. Then tell us what happened — we read everything.
Bickmore, T.W., & Picard, R.W. (2005). Establishing and maintaining long-term human-computer relationships. ACM Transactions on Computer-Human Interaction, 12(2), 293–327.
Nass, C., Moon, Y., Fogg, B.J., Reeves, B., & Dryer, D.C. (1995). Can computer personalities be human personalities? International Journal of Human-Computer Studies, 43(2), 223–239.
Reeves, B., & Nass, C. (1996). The Media Equation: How People Treat Computers, Television, and New Media Like Real People and Places. Cambridge University Press.
Spencer, L.M., & Spencer, S.M. (1993). Competence at Work: Models for Superior Performance. John Wiley & Sons.
Szathmári, E., Varga, Z., Molnár, A., Németh, G., Szabó, Z.P., & Kiss, O.E. (2024). Why do startups fail? A core competency deficit model. Frontiers in Psychology, 15, 1299135. doi.org/10.3389/fpsyg.2024.1299135
Tapus, A., Mataric, M.J., & Scassellati, B. (2008). Socially assistive robotics. IEEE Robotics & Automation Magazine, 14(1), 35–42.