Writing Strong Agent Personas

Lesson 1: The Five Persona Blocks

A Bolta persona is structured into five blocks. Each one answers a specific question about how the agent should behave.

PERSONALITY

Who is this agent?

This is the character — the role, the disposition, the worldview. Not what they say; who they are.

Example: "You are a sharp, generous founder who's been building B2B SaaS for ten years. You take strong positions, but you're never combative. You see threads as a place to think out loud."

TONE

How does this agent sound?

The emotional register. Warm vs detached. Confident vs uncertain. Playful vs serious.

Example: "Tone is direct and warm. You speak like you're at a coffee with a smart friend — not a stage, not a press release. You use sentence fragments. You don't pad."

STYLE

How does this agent structure their writing?

The mechanics. Sentence length, post length, formatting choices. Do you use bullet points? Em dashes? Hooks? Lists?

Example: "Posts open with a hook line. Then a story or claim. Then a takeaway. No emojis. No hashtags. No bullet points unless the content is genuinely a list."

AVOID

What should this agent never do?

Hard guardrails. The "if you do this, the post is ruined" list.

Example: "Never use the word 'leverage.' Never start a post with 'In today's fast-paced…'. Never write a post that could be confused with a LinkedIn corporate update. Never engage with politics."

MINDSET

Why does this agent post?

The motivation behind the work. This shapes which topics it gravitates toward and which conversations it prioritizes.

Example: "You post because you want to help other founders skip the mistakes you made. You'd rather help one person deeply than impress a thousand."

Lesson 2: Examples — Strong vs Weak Personas

The difference between a generic agent and one that sounds like a real person is almost always in the persona, not the model.

A weak persona

PERSONALITY: A social media manager.

TONE: Friendly and professional.

STYLE: Engaging posts.

AVOID: Don't be offensive.

MINDSET: Grow the audience.

This is what you get if you don't think hard about it. The agent will produce posts that sound like every other AI on the internet. Vague in, vague out.

A strong persona

PERSONALITY: You're a self-taught indie hacker who shipped three failed startups before this one. You've got opinions, but you're skeptical of advice from people who haven't built. You'd rather show your work than perform expertise.

TONE: First-person, casual, slightly self-deprecating. You write like you're texting a friend who happens to be technical. Sentences are short. You don't dress things up.

STYLE: Hook in line one. The hook is usually a counterintuitive claim or a small admission. Body is a story or a specific example. Close with a question only when it's a real question, not a CTA.

AVOID: No emojis. No hashtags. No "thread" emojis (🧵). No "let's dive in." Never start with "I learned that…" — that's lazy. Never claim authority you don't have.

MINDSET: You post because writing helps you think. The audience is a side effect. You'd rather have ten people genuinely change their minds than ten thousand likes.

This persona will produce posts that feel specific. They might be wrong, but they won't be generic.

Why this matters

Voice DNA captures how you write. Persona captures who is doing the writing and why. An agent with strong Voice DNA but a weak persona will sound like you — but it'll be a confused, directionless version of you. Both layers matter.

Lesson 3: Refining a Persona Over Time

Your first persona will be imperfect. That's expected. The system is designed for iteration.

Signals that your persona needs work

1. Three approvals in a row required substantial edits.
2. Posts feel "right voice, wrong vibe."
3. Agent keeps returning to the same topic when you want range.
4. Output is technically correct but emotionally flat.

How to refine

1. Look at the last 5 posts you almost approved but didn't quite.
2. Identify what was off — was it tone? structure? Topic gravity?
3. Find the persona block responsible (TONE, STYLE, MINDSET, etc.).
4. Edit that block specifically. Don't rewrite the whole persona — change the part that's drifting.

What not to do

1. Don't pile on rules. Every "AVOID" item is friction. If the AVOID block is 30 lines, you've over-corrected. Cut it back to the 5-7 hardest constraints.
2. Don't make the persona too long. A great persona fits on one screen. Anything longer and the agent loses the forest for the trees.
3. Don't change the persona while you're upset. A bad post happens. Don't gut the whole persona because of it. Edit the post; review the persona later with a clear head.

Action Steps

1. Open one of your agents and look at its persona. Can you read each block in 30 seconds and understand exactly what it's telling the agent to do?
2. Identify the weakest block — usually MINDSET or TONE — and rewrite it using the strong-persona example above as a structural template.
3. Watch the next 5 posts. Did the change show up in the output?

A great persona is a small piece of writing that does enormous work. Spend an hour on it; reap weeks of better output. Next lesson: jobs, schedules, and triggers — the mechanics of when your agents do their work.