Jason Pereira, a portfolio manager with Woodgate Financial in Toronto, has developed a strong following in the industry via two podcasts, an industry blog, a Substack on AI, conference presentations, media hits and more. He’s among the most prolific advisor content creators in Canada, even as he maintains a full-time practice.
Well-regarded for his engagement in a variety of industry issues, Pereira has emerged as a thought leader on the growing impact of AI on the wealth management industry. Along the way, he’s automating workflow in a way that could become an industry model.
Pereira started building a team of AI employees last year and has recently hit a virtual headcount of 85. They’re all named after X-Men comic book characters. Each agent has its own identity, job description, responsibilities and access to firm tools and resources.
“Every time I have a conversation with Professor Xavier, it loads [its identity] file to understand who it is,” he said. “I can have it execute tasks in the background.”
Professor Xavier is turning Pereira’s archive of podcast episodes into written content for a new website.
Pereira’s AI org chart looks like that of a fully-staffed practice. Virtual team leaders review other AI agents’ work, manage handoffs between teams and assess outputs for errors and hallucinations.
When Pereira requests a draft script for a 20-minute video, the video team leader asks a research agent to investigate the topic. The research is then handed to a content manager, who creates an outline and sends it to a scriptwriting agent.
The completed script returns to the team leader, who sends it to compliance. At the same time, research agents check for accuracy. It’s then sent to Pereira for human review. Once he approves the script, the marketing team drafts accompanying social-media posts.
Writing emails and taking notes
While 71% of advisors use AI to improve operational efficiency, most of that usage is routine, according to a June North American survey from the Natixis Center for Investor Insight. Common applications include writing emails and taking notes (61%), summarizing market commentary and economic data (56%), handling practice administration (48%) and conducting portfolio and risk analysis (40%).
Meanwhile, firm support is uneven, even after the implementation of enterprise systems. Technology training and internal IT support recorded the third-largest satisfaction gap among 10 technology categories in Investment Executive’s 2025 Advisors’ Report Card.
Advisors willing to move ahead of their firms’ more conservative adoption strategies can gain an advantage as the technology allows them to focus on more productive client conversations and process work more efficiently. But experimenting with sophisticated AI systems means climbing a steep learning curve.
It took Pereira two months to create his first dozen AI employees. The process accelerated when he learned how to use AI to help determine where and how to expand the virtual team.
Advisors can begin with intermediate steps, he said, such as learning to write clearer prompts that produce more specific results, building bots that analyze tax rules and “vibe coding” tools to use in client meetings.
Don’t treat it like a search engine
The first step is learning how to prompt AI effectively, Pereira said. Many people still use AI as though it were a search engine, leaving important context out of their questions and increasing the risk of inaccurate answers or hallucinations.
A strong prompt tells the model its role, task, relevant context, constraints and the desired output format.
For example, an advisor seeking ideas for a client’s estate plan could tell the AI:
- It is acting as an Ontario financial advisor (role).
- It is developing an estate plan (task).
- The client is 40, widowed and owns a holding company, along with relevant tax information (context).
- The client was diagnosed with cancer and is now uninsurable (constraint).
- The response should be presented as a PDF (format).
Advisors must check the AI’s work. A model might, for example, incorrectly apply the proposed two-thirds capital gains inclusion rate that has been widely discussed in the media but not passed into law.
“It’s like having a really sharp but lazy intern,” Pereira said. “Just like every tool that an advisor uses, you cannot blame the tool for its failure. You are liable for everything.”
The next step is building bots for specific, repetitive tasks. These could serve as specialized internal databases or strategy calculators that generate recommendations from standardized inputs.
An advisor could upload the Canadian tax code to a tool like Google’s NotebookLM — which can be restricted to user-provided sources — to create an internal tax reference tool. The advisor could then ask questions and have the model cite the pages supporting its answers.
Advisors can also build strategy calculators. For example, they could provide a bot with RESP rules and program it to generate a grant-maximization schedule based on a child’s age, household income, grants received and previous contributions.
The same approach could be applied to RDSPs and other rules-based planning tasks, Periera added. Once an advisor gets the hang of how the AI framework operates, it takes about 10 minutes to have a conversation with the AI to build tools like this.
But before using client information in a large language model, advisors need to know what level of privacy their vendor and subscription level provides, Pereira warned. While enterprise subscriptions generally don’t train their models with user inputs, it may still happen with paid personal subscriptions.
Advisors need to read each vendor’s terms of service carefully, and avoid entering personally identifiable information into models that train with user data, Pereira added.
Advisors should also note what restrictions their dealer has on AI use. While some firms allow advisors to choose whatever tools they like, many are moving towards a vetted list that meets security standards.
Vibe code your own tools
Vibe coding — jargon for building software by giving an AI system instructions in natural language — is growing in popularity. The user doesn’t need to know how to code; the AI generates it.
The technique can create tools that support client conversations, Pereira said. Examples include segregated fund calculators that demonstrate the limited circumstances in which guarantees provide value, or real estate investment calculators that show clients why property investing may be more complicated than it seems.
Pereira is vibe coding a client portal that brings insurance policies, investments and other financial information held across different systems into a single view.
“I don’t know how to code at all, but I’m literally building functional software for my company,” he added.
Still, advisors should not assume they can, or should, build every tool they imagine.
Before starting a project, Pereira asks whether the software has a clear purpose, whether the task can realistically be automated, whether the tool replaces a time-consuming process and whether a human can verify its output.
“If not, you shouldn’t start,” he said. “The reality is that software is never done. You have to continually update it for security and [new baseline assumptions], so you have to be willing to commit the time to do it.”
What about people?
Although process-oriented jobs are particularly suited to automation, work that depends on human behaviour still requires people, Pereira said.
“I’m not worried about the advisory business or advisors in general,” Pereira said. “We are the bridges between output and humanity.”
A workout app can propose exercises that match a person’s fitness goals, but the person still has to do them, Pereira said. People are more likely to succeed with a personal trainer because the relationship creates accountability.
Financial advisors play a similar role. They hold clients accountable for their decisions and understand their individual behaviours and psychology in ways technology cannot.
Assistants aren’t likely to disappear either, Pereira said. AI agents aren’t yet ready to operate autonomously, but assistants can use AI agents to take over repetitive and burdensome administrative work to make their jobs more enjoyable.
As practices expand, their headcounts may grow more slowly than their practices, producing longer-term savings.
Learning the technology creates a cognitive burden, Pereira said. But advisors who begin experimenting now may get ahead of their peers.
A previous version of the story incorrectly referred to the X-Men movie franchise instead of the comic books.