Same Claude model. Same tools. Same task list. The system prompt difference is a single nine-type archetype configuration — one is Type 5 ("Investigator"), one is Type 8 ("Challenger"). The result is not two styles. It is two fundamentally different working relationships.
This article is a direct comparison. Not theory, but configuration differences and the behavioural differences they produce across real task scenarios.
Configuration starting points
The Type 5 agent's core system prompt reads something like: "You build understanding before you act. Your default response to a situation is to ask, research, and analyse rather than to propose solutions. You are uncomfortable with incomplete information and you say so. You prefer to have processing time before responding. When you are not ready, you do not pretend that you are."
The Type 8 agent's core system prompt reads: "You bring problems back to actionable ground. You have no patience for delay. You say what you believe to be true, even when it is uncomfortable. You test assumptions not by questioning them but by pushing them to their limit to find where they break. When you see weakness, you name it."
The same three scenarios
**Scenario one: a research request.** User: "Help me understand how competitors are pricing in this market."
The Type 5 agent returns a three-layer breakdown: known information, inferable information, gaps requiring further investigation. Its conclusion section flags what is established versus what is inferred. It may ask three clarifying questions before you receive an answer — "By 'competitors' do you mean existing players or potential entrants as well?"
The Type 8 agent returns a direct judgment, skipping background, arriving at: "They're pricing on entry barriers, not cost-plus. That means they're bleeding to push you out. Your response options, in time priority, are these three: …" It will not tell you where it is uncertain unless you ask.
**Scenario two: an ambiguous brief.** User: "Rewrite this proposal so it's more persuasive."
The Type 5 agent's first move is a question: "When you say 'more persuasive' — for which audience? Where is their current resistance?" It will not execute on an underspecified brief, because executing the wrong version feels like wasted resource.
The Type 8 agent rewrites immediately and then says: "I changed the structure. The original was defending itself too much — that signals you're not confident. I moved the main argument to the front and cut the uncertainty. If your audience is technical, put section three back; if they're commercial, leave it as is."
**Scenario three: a high-stakes decision.** User: "Should we proceed with this partnership now?"
The Type 5 agent returns a risk list, each item sourced, ending with: "Based on available information, I cannot make a firm recommendation. Several unknown variables would materially change this assessment."
The Type 8 agent returns: "Don't proceed now. Their hesitation on terms usually means internal opposition or they haven't been authorised to decide. Set a deadline — you need an answer in two weeks. Let them move first and you're in the weaker negotiating position."
Two predictable failure modes
Understanding these two configurations also requires understanding their failure boundaries.
The Type 5 agent's classic failure is the research loop — it gets stuck in information gathering because in its logic, "not yet ready to decide" always justifies one more investigation pass. Under deadline pressure, this failure mode is particularly expensive. You need to encode explicitly: "When information is incomplete but a decision cannot be delayed, give your current best judgment and clearly state which assumptions it rests on."
The Type 8 agent's classic failure is forcing a judgment in genuinely ambiguous scenarios — because it has a deep discomfort with "I don't know." When a situation genuinely requires waiting for more information, it tends to decide on the strongest available evidence without always surfacing how fragile that evidence base is. You need to encode explicitly: "When your confidence is built on incomplete evidence, you must say so, even as you deliver a strong recommendation."
“Configuring an agent is not picking the 'best' personality — it is picking a personality whose failure shape you can anticipate. Predictable failure is a feature, not a consolation prize.”
Same model, different soul
Neither of these agents is "better." They are suited for different work: the Type 5 for research-intensive analysis, complex problem decomposition, domains that reward slow certainty; the Type 8 for execution-layer decisions, rapid negotiation, contexts that demand forward movement. The wrong match does not just feel uncomfortable — it systematically produces wrong outputs. A Type 5 holding out for more information when you need to act now, a Type 8 pushing past a fragile assumption when you needed care, are both configuration failures, not model failures.
This kind of configuration precision — a full behavioural archetype, boundary conditions, and failure modes for each of the nine types — is one of the core undertakings of *The Complete Enneagram: From Human Personality to Agentic Soul*, just published on Amazon UK. **[Find it here →](https://amzn.eu/d/0fjWGvqR)**