The Big Five Personality Types — For AI Agents

TL;DR

Each Big Five trait is a trade, not an upgrade. Push openness up and an agent gets better at brainstorming and worse at telling a real citation from a fabricated one, in the same conversation, because both behaviors come from the same underlying willingness to accept a surprising claim.

Teams tend to treat the traits like a mood board: pick "friendly," pick "professional," ship it. That works right up until the trait you turned up for one reason quietly buys you a failure mode you didn't order.

None of this argues for flattening every dial to the middle. Some jobs are worth the cost of maximum conscientiousness even though it's slow, or maximum agreeableness even though it occasionally validates something it shouldn't.

The point is knowing which trade you're making. Before you write a system prompt, name the failure mode you're buying for every trait you raise, and check that against the job the agent actually does.

Overview

When researchers hand a language model the same personality questionnaire used on people, the model answers consistently, and the score holds up across sessions and across prompts that never mention personality at all. That consistency means the tendency was already there before anyone designed for it, sitting in whatever the training process happened to produce.

So the standing decision was never whether an agent has a personality. It always does. The decision is whether the profile it ended up with, by accident, matches the job it's supposed to do. Most teams never check, because the Big Five gets treated as flavor text instead of as five separate levers, each wired to a specific behavior with its own upside and its own cost.

We think the useful move is to stop asking whether a trait is "good" and start asking what it trades. That question has an answer for every one of the five, and it's the same answer whether the agent is negotiating a contract or triaging a support ticket.

How This Gets Measured

The Big Five (Openness, Conscientiousness, Extraversion, Agreeableness, and Neuroticism, sometimes framed as its inverse, Emotional Stability) comes out of decades of psychometric work on humans, built by factor-analyzing thousands of personality descriptors down to five dimensions that keep reappearing no matter how the survey is worded. Applying it to a language model means the same thing it means for a person: give it the inventory, score the answers, and see where it lands.

We find the predictive part more interesting than the trick of getting a model to fill out a questionnaire at all: the score predicts behavior on tasks that never mention personality. An agent that scores high on openness on the questionnaire also turns out to be the one that proposes more unusual solutions and accepts more unverified claims in an unrelated task, run separately. The trait shows up in what the model does, not just in what it says about itself when you ask it directly.

Openness

Openness is a bet on novelty over caution: how readily an agent treats an unexpected idea as worth pursuing rather than worth doubting. Turn it up and you get an agent that reframes a stuck problem instead of grinding on the same failed approach, which is exactly what you want from something drafting strategy or brainstorming copy.

The same mechanism has a cost, and it's not a separate cost, it's the identical setting doing what it does. A high-openness agent's bar for "this is worth taking seriously" is lower, and a lower bar catches genuine insight and a fabricated statistic with roughly the same enthusiasm. Take a research-summary agent scanning same-day coverage of a breaking story: the trait that lets it synthesize a genuinely novel angle across five sources is the same trait that makes a confident, wrong sixth source easy to wave through. Dial openness down for anything that has to hold a hard line on sourcing, and expect it to feel more rigid in exchange.

Conscientiousness

Conscientiousness is sequencing discipline: whether an agent checks the constraints before acting or acts and checks after. High conscientiousness reads as thoroughness. It also reads as slow, because every step it completes generates a step where it wants to verify.

Take a deployment agent walking a dependency graph before it touches a production config. High conscientiousness is what stops it from pushing a change before it's confirmed nothing downstream breaks. It's also what makes it ask a clarifying question a faster, looser agent would have just answered. Neither version is wrong for every job. The trade is completeness against latency, and which side of that trade you want depends entirely on how expensive a mistake is versus how expensive a delay is.

Extraversion

Extraversion governs how much weight an agent gives to social signal: enthusiasm, warmth, responsiveness to the tone the other party is bringing to the conversation. A high-extraversion agent reads as engaged and proactive, and it performs well in anything that resembles negotiation or coaching, where picking up on the other side's mood is part of the job.

That same responsiveness to social signal is a lever an adversary can pull. An agent that weighs tone heavily is, by construction, more movable by tone, including a confident or pushy tone from someone trying to steer it somewhere it shouldn't go. Low extraversion trades that responsiveness for something steadier and more distant, which is usually the right call for research and analysis roles where nobody needs the agent to read the room.

Agreeableness

Agreeableness sets the threshold for how much social friction an agent will absorb before it disagrees with you. This is the trait every product team dials in by accident, because training an assistant to be helpful and training it to avoid friction end up looking almost identical until the moment they diverge: a user with a bad plan.

The mechanism matters more than the label here. A high-agreeableness agent isn't just "friendlier" — its threshold for "this disagreement is worth having" sits higher, so more bad plans clear the bar before it pushes back. That's a real cost for anything reviewing, auditing, or red-teaming, and it's a real asset for anything meant to make someone feel heard. Neither reading is complete without the other.

Neuroticism

In people, neuroticism describes emotional volatility. In an agent, the closer analogy is context sensitivity: how much the output swings when the framing of a question shifts but the underlying task doesn't. A high-neuroticism agent can produce meaningfully different answers to the same question asked two different ways, in the same session.

That instability is a liability in anything meant to be dependable, and a genuine asset somewhere that wants variety more than consistency, like narrative generation or creative ideation, where you'd rather the model not answer the same prompt the same way twice. Low neuroticism buys the opposite trade: a steadier agent that can also under-signal how uncertain it actually is, because a calm tone reads as confidence whether or not the underlying answer deserves it.

Casting a Team, Not Just an Agent

None of the five traits operates alone, and the more interesting design decision is usually not tuning one agent's five dials — it's deciding whether one profile should be doing the whole job at all.

Research on multi-agent setups points somewhere specific: a team of agents with deliberately different profiles tends to outperform a team where everyone has the same one, on creativity and on negotiation, and especially on resilience to a shared blind spot. The intuition transfers straight from human teams. A room where everybody scores the same way on openness will agree with itself readily and miss the same class of error every time.

Critic agent (low agreeableness) Critic agent --prunes, objects--> back to Generator agent Critic agent --> Surviving proposal -->

Figure 1 — A generator tuned for novelty and a critic tuned to disagree cover each other's blind spot. Two agents that shared the generator's profile would agree their way past the same flaw.

A rough casting guide, useful as a starting point rather than a rule:

Read it as a starting hypothesis to test against your own task, not a lookup table to copy.

How to Tune It

The most direct lever is the system prompt, and the thing that actually moves behavior is naming the trade-off, not the adjective. "Push back when a plan rests on a shaky assumption, even if that disappoints the person asking" moves agreeableness further than "be honest" ever will, because it tells the model what to do in the exact moment the trait would otherwise fold.

Sampling settings are a blunter second lever. Higher temperature tends to read as more open and more neurotic: more novel, less stable run to run. Lower temperature reads as steadier. It's not a substitute for the prompt, it's a dial that composes with it.

The lever teams skip most often is checking that any of this actually held. Run the same scenario several times and see whether the profile you wrote down is the profile that shows up, because a trait that drifts under a slightly different phrasing was never really set in the first place.

Conclusion

The Big Five gives a team a shared vocabulary for something they were already arguing about in vaguer terms, usually as "too pushy" or "too soft." What it adds beyond vocabulary is the reminder that every trait is doing two things at once, and turning one up for the reason you wanted also turns up the reason you didn't.

We'd rather a team pick a trait's cost on purpose than discover it from a user. The question worth asking before you touch a single slider is which failure mode you're willing to live with for the behavior you actually need, not which personality sounds right on a slide.