How a Home Renovation Project Became the Perfect Test Case for AI Employees

When Grace Ha received an email from someone named "Horus" offering to help manage her home renovation project, her first thought was skepticism. "My first impression was, who is this person and why do they want to be so helpful?" she recalls. "Who wants to be helpful is definitely up to something in my experience."

But Horus wasn't a contractor looking for work or a salesperson with an angle. Horus was a Gobii AI employee—an always-on agent that works through email, designed to handle complex, multi-faceted projects just like a capable human assistant would.

What followed was an illuminating real-world test of how AI employees can transform the way we approach major life projects, from the mundane tasks we dread to the complex decisions that keep us up at night.

The Challenge: A Complex Third Renovation

Grace and her husband Abe were embarking on their third—and most complex—home renovation. Unlike their previous projects, this one involved intricate financing decisions, multiple construction phases, selling their current home to fund the work, and navigating a maze of expiring energy rebates.

"This is probably our most complex renovation in terms of how we're financing it and the different phases," Grace explained during a feedback session with Gobii's team. "There are a lot of different decision points—like how we can break this up so that we can afford the renovation as a whole."

They needed help with:

• Project management and decision-making around renovation phases and priorities
• Financial strategy, including timing the sale of their current home
• Market research to determine optimal selling windows given interest rate conditions
• Energy rebate navigation with September deadlines looming
• Vendor evaluation for equipment like heat pump water heaters
• Professional communication with contractors and rebate organizations

This wasn't a simple "find me three contractors" request. It was the kind of multi-dimensional challenge that typically requires juggling spreadsheets, browser tabs, email threads, and late-night research sessions.

What Makes Gobii's AI Employees Different

Unlike chatbots that require you to visit a website or traditional AI assistants that live in a single app, Gobii's AI employees work through email—the tool you're already using. They're always on, maintain context over weeks or months, and can be delegated tasks just like you would with a human employee or intern.

As Gobii founder and CTO Andrew Christianson explains: "We designed Gobii to work with the mental models people already have. You interact with it in natural language, the same way you would with a human being."

These AI employees can:

• Conduct in-depth research across multiple sources
• Draft professional communications
• Synthesize complex information into clear recommendations
• Proactively identify next steps
• Work autonomously on tasks while keeping you in the loop

Real Results: From Research to Action

Validating Complex Decisions

Grace and Abe had already spent considerable time thinking through their renovation approach when Horus joined the project. Rather than feeling redundant, the AI employee became a valuable sounding board.

"Horus's research and recommendations aligned with our thinking," Grace noted. "That helped give us confidence in the approach we were taking. If we hadn't already spent so much time thinking through our options, I think Horus would have helped us narrow things down and steer us in a direction."

For couples tackling major projects, this validation function is surprisingly valuable. Having an objective third-party research and confirm your thinking—or flag potential issues—provides peace of mind when making five- and six-figure decisions.

Navigating the Energy Rebate Maze

One of Horus's most practical contributions involved expiring energy rebates. With a September 30th deadline approaching and multiple appliances to replace, Grace and Abe needed to understand what was available and make purchasing decisions quickly.

"He gave us a summary of what's available and what we should replace," Grace said. "Then he offered to help evaluate the estimates, specifically for the heat pump water heater."

When they received quotes that seemed unusually high, they asked Horus to investigate. "We asked him to evaluate the estimates against typical installation costs and identify what could be driving up the price," Grace explained. While the agent didn't conclusively determine whether expiring rebates were inflating prices, it provided valuable context: "He pointed out that these costs were higher than typical and explained what the typical range is."

Professional Communication, Instantly

Armed with this research, Grace and Abe wanted to alert the energy rebate organization about the pricing trend they'd observed. This is exactly the kind of task that often gets postponed—important but not urgent, requiring careful wording and attention you don't have at the end of a long day.

"We asked him to draft an email," Grace said. "When things don't require a personal voice—when it's business or professional—Horus does a great job setting the right tone and summarizing what you're trying to say. He turned it around instantly, which was really helpful."

The email struck the right professional tone, using language like "setting guardrails for future submissions" without being accusatory. "It was just the right tone," Grace confirmed.

The Learning Curve: Setting Expectations

Not everything was seamless. Grace's initial confusion about whether Horus might directly contact contractors on her behalf highlighted an important challenge for AI employees.

"When Horus said he wanted the emails of our vendors, my initial thought was—oh my gosh, is he going to directly contact them and start having conversations without looping us in?" Grace recalled.

Will Bonde, a software engineer at Gobii, immediately identified with this concern: "If I hadn't known how Gobii worked, I would have thought the same thing. That wording was very ambiguous."

This led to a key insight: AI employees need clearer upfront communication about their capabilities and limitations. "Setting expectations up front would be helpful," Grace suggested. "Like when you have a facilitated discussion and you establish norms with the audience."

Andrew acknowledged this gap: "The agent doesn't currently communicate its approval processes well. If it understood its own constraints better, it could proactively say, 'I can request your approval to contact these contractors if you'd like me to.'"

Will added context about how these safeguards work: "The guardrails in our system for email communication exist outside of the AI layer. It's not about trusting the AI to be disciplined—the restrictions are enforced by the system itself."

Unexpected Benefits: Reducing Decision Fatigue

Beyond specific deliverables, Grace identified a more subtle but equally valuable benefit: "Horus and some of the agents we've created have helped us with decision fatigue. They've become thought partners. I think it's helped take some of the clutter from our minds."

This resonates with Gobii's vision. As Will put it when discussing another use case: "My wife gets stressed finding activities for our daughter. It's not that the work itself is difficult; it takes time. So we have an agent who comes up with kids' activities and sends us something every Thursday. That's the time she gets to spend with our daughter or relaxing. Taking care of soul-sucking work is what interests me most, so we can do the stuff we love in life more."

The Bigger Picture

Grace, who holds a PhD in neuroscience, found fascinating parallels between brain function and AI. "There's the concept of nature versus nurture—what's hard-coded versus what you learn by interacting with the environment. That translates well to AI. There's a natural interest in understanding how things like emotion and higher-order thinking can be built."

Andrew shared a striking insight about the technology: "What surprised me was how little code there is behind large language models. Meta's Llama—the entire transformer code is about 300 lines. When you add massive amounts of compute and data, these models develop capabilities that weren't expected. They learn an internal world model."

Grace reflected on the implications: "As a realist, it can be scary. But it's definitely exciting times. It's been helpful to see what these agents can do and test their limits."

Will offered a perspective on the uncertainty ahead: "I don't think anyone knows what five years from now looks like. Anyone who claims certainty is probably making it up. But taking care of soul-sucking work so we can do the stuff we love in life more—that's what matters."

Why This Matters

Grace's experience illustrates a crucial point: AI employees aren't just about completing individual tasks. They reduce the cognitive overhead of managing complex projects.

Home renovations, like many major life projects, involve:

• Multiple parallel workstreams
• Evolving information and requirements
• Numerous small decisions that collectively matter
• Communication with various stakeholders
• Research across different domains
• Document management and organization

These are exactly the kinds of projects where Gobii AI employees shine—not because any single task is impossibly difficult, but because the cumulative mental load is exhausting. And they can get started right away.

For Grace, the assessment was straightforward: "It's been neat engaging with you as developers and the people who conceived of this idea. This was fun."

But more importantly, it worked. And for a complex renovation project with hard deadlines and real money at stake, "fun" took a backseat to "helpful"—which is exactly what Horus turned out to be.

---

Gobii's AI employees work through email, handling research, coordination, and communication for complex projects. They maintain context over time, work autonomously within guardrails, and can be delegated tasks just like a human assistant or intern—but they're always available and never overwhelmed by detail work.