Behind the Scenes: How We Built AI Assistants That Actually Feel Human

When we started building Vasara’s AI Assistants, one problem kept us up at night: every AI conversation felt like talking to a robot.

You ask ChatGPT to write in your brand voice – it gives you generic corporate-speak. You tell it to be “more casual,” and it suddenly sounds like it raided a teenager’s text messages. The personality feels forced, inconsistent, and fake.

We realized the issue wasn’t that AI can’t sound human. It’s that everyone was trying to make AI sound human without first understanding what actually makes humans sound human.

So we asked a different question: What if we built AI assistants the same way companies build human teams?

The Personality Problem: Why Most AI Feels Robotic

When you hire a copywriter, you’re not just hiring someone who can write – you’re hiring someone whose style, personality, and professional approach fit your culture. They learn your terminology, absorb your values, and adapt to your way of communicating.

Most AI assistants skip this step entirely. They’re built to be generically helpful – a one-size-fits-none approach that works for no one particularly well.

We found three major gaps:

• No Shared Vocabulary: Human teams use internal shorthand and industry-specific language. AI uses sterile, generic terms.

• Missing Communications Alignment: A scrappy startup and a polished agency speak very differently. Most AI systems doesn’t take those differences into account.

• No Cultural Context: Humans absorb culture through daily interactions. AI gets a few prompts and hopes for the best.

Our hypothesis: Teach AI assistants to absorb company culture the way new hires do.

Building Personality From Data, Not Prompts

Instead of forcing personality through clever prompts, we reverse-engineered it from real data.

Phase 1: Public Brand Voice Analysis

Every company already has a voice – embedded in your website, social media, blogs, and marketing. This is your external – polished, intentional – persona.

Our Context Engine analyzes:

• Vocabulary patterns: “Customers” or “clients”? “Purchase” or “buy”?

• Sentence structure: Short and punchy, or long and explanatory?

• Tone indicators: Formal, casual, witty, authoritative?

• Value expressions: How do you talk about quality, service, innovation?

This gives us the public voice – what brand customers expect to hear.

Phase 2: Internal Communication Learning (Where the Magic Happens)

Public voice is only half the story. The real culture shows up inside – emails, Slack threads, meeting notes. And this part is more complex and tricky compared to the public one.

Internal communication reveals:

• True working vocabulary: The terms your team actually uses

• Decision-making language: How you discuss priorities and solutions

• Cultural values in action: Not just what you say you value, but how you live it

• Personality quirks – The small things that make your team distinct

A few examples: A marketing team called urgent tasks "code red" instead of "high priority." Their AI vocabulary adopted the phrase, and now when deadlines loom, the AI says "This looks like a code red situation" – and the team immediately gets the urgency. A tech startup used "parking lot" for ideas to revisit later. Their AI learned this and now suggests "Should we park this for now?" when conversations drift off-topic.

This step requires permissions, privacy protection, and careful handling – but it’s what makes our AI stop sounding like a press release and start sounding like your teammate.

From Generic to Genuine: Creating Real Personas

After building the technical foundation, we faced a uniquely human challenge: What should each AI assistant actually be like as a person?

We could have assigned random personalities. Instead, we studied hundreds of real professionals in each field.

Juno emerged from research on business strategists – a composite of the analytical, San Jose-area professionals who've spent years helping companies scale. Plume came to life through studying Brooklyn's creative scene – the innovative copywriters who thrive in one of the world's most competitive markets.

The difference is profound. These aren't chatbots with surface-level personalities. They're AI teammates with authentic professional voices, communication styles, and expertise that feels genuinely human.

Skills vs. Conversations: Knowing When to Switch

Here’s something most AI companies miss: people don’t always want to chat with AI. Sometimes they want speed.

• Skills: Our efficiency engines. Purpose-built for repeatable tasks, like creating a Facebook post 5× faster because the Skill already knows your brand voice, platform requirements, and preferences.

• Conversations: Flexible, dynamic, great for brainstorming or handling surprises.

The elegant part? If you’re chatting with Plume about Instagram content, she’ll naturally suggest launching the Instagram Reel Script Skill. Same personality, just a faster tool.

The Context Engine: Memory That Feels Human

One of the hardest technical problems in building AI assistants is memory. Humans naturally remember preferences, feedback, and context, while most AI systems forget everything the moment you close the chat.

Our Context Engine changes that by managing three distinct types of memory in harmony. It holds on to semantic memory, which stores the key facts about your business – your products, services, audience, and other essential details. It also retains episodic memory, which keeps track of specific interactions, noting what worked, what didn’t, and the feedback you’ve provided along the way. Finally, it develops procedural memory, which understands how you like things done – your style, processes, and workflows.

By weaving these memory types together, the Context Engine keeps what matters most, filters out the noise, and allows your AI assistant to feel like it has been working alongside you for years.

Making It Seamless

Through secure integrations with Gmail, Facebook Insights, Google Calendar, and more, assistants pull real-time data without you lifting a finger.

When Juno mentions your website traffic spike, he’s looking at actual analytics. When Echo recommends posting times, she’s reading your real social performance. The tech is invisible. The value is instant.

Our Never-Ending Mission

We believe great AI assistants should feel less like software and more like trusted teammates. Getting there means constantly pushing the boundaries of what's possible.

Right now, we're perfecting the art of seamless transitions – how an assistant naturally shifts from casual conversation to specialized Skills without missing a beat. Think of it like a colleague who instinctively knows when to switch from brainstorming to execution mode.

We're also working on privacy-first learning, where AI gains deep business context while maintaining enterprise-grade security. Your data stays yours, but your assistants get smarter.

Most importantly, we're eliminating the complexity. Our goal is AI teammates that understand your business as deeply as veteran employees, but onboard as easily as any modern app.

The vision: AI collaboration so intuitive, you'll forget there's technology involved at all.

The Future of AI Assistants

Other companies are racing to make AI smarter. We're making it more collaborative.

The goal isn't an AI that does everything – it's specialized AI teammates that excel in their roles while seamlessly working together. Echo doesn't just post content; she understands your brand voice. Plume doesn't just write copy; she captures your authentic tone. Juno doesn't just analyze data; he thinks like your strategic partner.

Because the best AI assistants won't just sound human. They'll sound like the professionals you'd actually want to hire.