The Peel with Turner Novak: Inside Canada's Fastest Growing AI Company | Spellbook, Scott Stevenson

Date: March 12, 2026
Guest: Scott Stevenson (CEO & Co-founder, Spellbook)
Host: Turner Novak

Episode Overview

In this episode, Turner Novak interviews Scott Stevenson, CEO and co-founder of Spellbook, which has rapidly become Canada’s fastest-growing AI company. Spellbook is transforming how legal teams draft, review, and negotiate contracts by embedding AI directly into lawyers' existing workflows—particularly via a Microsoft Word plugin. The conversation uncovers Spellbook’s origin story, how the legal tech industry is evolving under the impact of generative AI, lessons learned from product-market fit struggles, and insights into the future of AI-powered work.

Key Discussion Points & Insights

1. What is Spellbook? (00:22-01:51)

• Spellbook’s Core Functionality:

  • “We're basically a cursor for contracts. So an AI copilot for contract review and drafting.” (00:29–00:32, Scott)
  • Built first as a Microsoft Word plugin, it’s now used by 4,000 customers in 80 countries.
  • Focuses on the “commercial legal work” niche: drafting/reviewing contracts (leases, sales agreements, VC financing docs) by both law firms and in-house legal teams.

• Bottoms-Up Growth Strategy:

  • Spellbook’s go-to-market departs from traditional legal tech: “We really sell bottom up to the lawyers and the contract managers who are using the software and, you know, kind of organically expand upwards from there.” (02:12–02:34, Scott)

2. Product Deep Dive: Why Word? What’s Unique? (03:13-06:40; 10:20-12:12)

• Word Plugin Advantage:

  • Inspired by Github Copilot for coders: “The core of the product sits on top of Microsoft Word, which is where most lawyers are doing their drafting and reviewing work.” (03:18–03:30, Scott)
  • Word is still the dominant legal drafting tool due to its established formatting and collaboration features; integration there removes friction.
  • Building the plugin itself is “just a web page...connects to Word API...hardest part is manipulation of the Word document” given the file format’s age and quirks—like embedded spreadsheets. (10:31–11:28, Scott)

• Spellbook’s Approach:

  • The goal isn’t full automation: “Our idea of a great product for lawyers is that it should be like an electric bicycle...they're still steering, they're still pedaling...but now they can get up over the hills a lot easier.” (03:49–04:29, Scott)

3. Major Use Cases & Features (06:40–09:22)

• Autocomplete & Drafting:

  • Initially focused on autocomplete (like code copilots), now does much more.

• Contract Review:

  • “The biggest thing...is contract reviews. So you can take any contract, say like a lease, instantly review it for risks and issues, and it will learn over time what you tend to flag and what you don't. So it gets better and better.” (06:46–07:19, Scott)
  • Review process is subjective (like YouTube recommendations); suggestions are tailored to users' habits.

• Redlining & Playbooks:

  • Automatically redlines documents using “suggested edits” with Word’s track changes.
  • Playbooks allow in-house legal teams to standardize negotiation rules (“maybe they have 20-40 rules that dictate...what they allow or won’t...it will kind of automatically do that negotiation”). (08:22–09:22, Scott)

4. How the Legal Market Works & AI's Impact (13:49–19:21)

• AI’s Transformative Potential in Law:

  • “Large language models...were kind of like the spreadsheet moment for lawyers.” (14:17–14:22, Scott)
  • Unlike finance, which automated with spreadsheets and software decades ago, law hasn’t automated due to reliance on unstructured text—until LLMs arrived.

• Legal Software Stack (Pre-AI):

  • "Word, Outlook...DocuSign...that was essentially the lawyer stack." (17:21–18:18, Scott & Turner)
  • $30 trillion in economic value flows through contracts every year—ripe for efficiency gains (18:31–18:39, Scott).

• Standardization vs. Customization:

  • Templates/standard docs help, but “most of law has been surprisingly resistant to standardization because the deals people want to make are all kind of unique and bespoke.” (19:28–20:42, Scott)

5. Why Not Just Use ChatGPT? Custom AI vs. Vertical Integration (21:08–26:32)

• Why Not Use ChatGPT Directly?

  • Lawyers rarely draft from scratch and want to work off trusted precedents: “If they go to ChatGPT...then they have to review every single word of that and make sure they understand it completely. That's very difficult.” (21:13–21:31, Scott)
  • Spellbook allows surgical edits to existing, familiar contracts and offers unique market comparisons (“compare this to the average commercial lease in Manhattan and tell me what's not normal”—22:55–23:07).

• Aggregated Data & Customer Privacy:

  • Clients can opt in to anonymously aggregate contract stats (e.g., lease terms), which generates valuable market insights while preserving confidentiality.
  • Fine-tuning models was hyped—but didn’t pan out; “RAG (Retrieval Augmented Generation) is actually really, really good and actually superior to fine-tuning.” (26:07–27:16, Scott)

• On Fine-Tuning vs. RAG (Retrieval Augmented Generation):

  • “Fine-tuning was this big meme…but it actually works pretty terribly for a whole bunch of reasons.” (25:14–25:32, Scott)
  • RAG allows users to get verifiable, filterable, and privacy-safe results pulled from real data—not the “evolutionary instincts” (aka hallucinations) of a giant model. (27:10–28:18)

6. Platform Moats & Product Defensibility (32:28–37:08)

• Why Won’t OpenAI or Anthropic Just Build It?

  • Differentiation is necessary, but “the pendulum swung too far in the direction of differentiation rather than customer value.” (32:53–33:14, Scott)
  • Spellbook stays “two years ahead at all times in terms of delivering state of the art experiences.”
  • Workflow integration (i.e., being “the toaster product” for contracts) is a core differentiator: “If we can just do that one thing well...you make so many decisions differently that would never make sense for ChatGPT to make.” (35:08–35:38, Scott)

• Network Effects:

  • Aggregated contract data provides a powerful advantage, especially across countless geographies/industries that require deep, context-specific insight. (36:12–37:08)

7. Sales Model: Bottoms-Up vs. Top-Down in Legal (37:33–42:26)

• Traditional Legaltech Sells Top-Down (Harvey, Lagora, etc.) to Large Firms:

  • Top-down sales focus on innovation teams, which often leads to slow decision cycles and complicated feature requests.
  • “At a lot of these large law firms, they operate on an hourly billing model and just decreasing...billable hours is not a positive incentive.” (38:55–39:14, Scott)
  • Client portals are often requested for "innovation show-and-tell," but lawyers/clients didn’t actually want to use them. (39:56–40:21)

• Spellbook’s Bottom-Up Motion:

  • Product is adopted at the end-user level (“lawyers and contract managers”), with minimal onboarding friction, leading to high retention and organic expansion (NDR >130%).

8. The Landscape of Legal AI Companies (42:26–46:20)

• Harvey & Lagora: Focused on top-down deployment for major law firms—more like “ChatGPT for law firms,” broad in scope.
• Spellbook: Dominant in enterprise in-house legal teams, where productivity gains are meaningful (e.g., Dropbox, eBay are customers).
• Others Mentioned:
  • Even Up (personal litigation),
  • Hebia (shifted to finance),
  • Sandstone (enterprise legal).
• Rapid Consolidation: The need for speed in market/feature development means smaller entrants are being acquired or shutting down. (46:24–46:56)

9. Adapting to the Pace of AI Development (47:22–54:04)

• Ability to Rapidly Switch to New Models Is Key

  • “There’s no time to master anything. Every six months, a new tool or a new model comes out…no point in sharpening the ax when the chainsaw is coming out tomorrow.” (47:33–49:36, Scott)
  • Product and engineering culture must embrace continual iteration and scrapping of legacy code.

• Building for Tomorrow’s Tech

  • Spellbook often starts building features the models can’t yet support, anticipating those capabilities will catch up within six months. (51:15–52:33)
  • “If you build something that's achievable today, it's not going to be that impressive in six months.” (52:26–52:33, Scott)

• Agents and Autonomous Workflows

  • Spellbook’s next horizon is “AI agents working in the background all the time like an actual employee...I think that is going to be like a 10x for AI and agents.” (55:43–56:59)

10. The Human Element vs. Automation (58:54–61:26)

• What Should Humans Still Do?
  • On writing and creative work: “If it wasn't worth the time for this person to write this, then it's not worth the time for me to read it. … In a way, I think…the fact that…someone's willing to actually sit down and write something themselves indicates that they thought it was important enough to invest that time.” (59:18–59:58, Scott)
  • Legal work (contracts, code) is not about creativity, which is why AI is especially effective here.

Spellbook’s Founding Story: Iteration, Adversity, and Breakthrough (61:26–80:19)

• Scott’s First Startup:

  • Invented an electronic music instrument, learned the pain of expensive legal bills. “One day, we got a 10k legal bill by surprise…that was an enormous amount of money.” (63:19–64:03, Scott)

• Pivot to Legal Tech:

  • Tried “smart contracts” (blockchain), then built Rally (template-based contract drafting).
  • Persistent lack of product-market fit—thus, kept team lean, running over 100 landing page experiments: “We would drive traffic...and we would see what's the cost per conversion...we literally tracked that over 100 landing pages...” (70:21–71:20, Scott)

• Breakout with Spellbook’s AI-based Product:

  • First to market with GPT for lawyers, summer 2022, before ChatGPT launched.
  • “Within three months we had 30,000 waitlist signups. Within three months we had more revenue from that product than the other three years we had selling everything else.” (73:28–74:23, Scott)

Fundraising, Growth & Working With Keith Rabois (81:10–90:41)

• Achieving True Product-Market Fit

  • “When we hit that moment, our salespeople's calendars were just completely blocked...every day of the week they would have eight sales meetings” (75:11–75:47, Scott)
  • Once the AI product resonated, scaling was an obvious next step.

• Series B Fundraise:

  • Announced with a viral growth chart tweet (“I tweeted our growth chart…my email…I still have emails I haven’t responded to from that moment from funds who reached out.” – 83:42–85:56, Scott)
  • Noted differences between NYC (quantitative, spreadsheet-driven) and SF investors (more vision-focused).

• Lessons From Keith Rabois (KV):

  • “He obsesses about performance and like what best in class people look like…[he’s] a nuclear grade communicator.” (87:35–90:13, Scott)
  • Importance of repeated, clear, simple messaging: “To do it, you have to have a really simple message and you have to repeat it a lot.”

Scott's Personal AI Stack (91:00–94:35)

• Tools He Uses: Cursor, Claude Code for prototyping.
• Favorite: Twin
  • “It’s on the surface, very simple…generalized agent formula, really right…you can say, I want to build an agent. You build an agent by prompting—you don’t have to code.” (91:32–91:49)
  • Used for tasks like scanning Twitter for Canadian AI talent; aggregating product feedback across channels.
  • Key: Agents push results into Slack channels for better usage.

Notable Quotes & Moments

On Spellbook’s approach:

• “Our idea of a great product for lawyers is...an electric bicycle. They're still steering, they're still pedaling...but now they can get up over the hills a lot easier.” (03:49–04:29, Scott)

On top-down vs. bottom-up in legal AI:

• “Lawyers make a lot of money, especially in these large firms, from billable hours. And so what you found was ... what the committee's most concerned about is how do we advertise this to our clients?” (39:14–39:56, Scott)

On the hype around fine-tuning AI models:

• “I do not know of a single [legal-specific fine-tuned] model that is still in use today at any of the major application providers.” (26:07–26:32, Scott)

On product-market fit:

• “Our view of product market fit is basically the customer is pulling the product out of your hands faster than you can keep up with…” (68:48–68:53, Scott)

On culture and engineering in AI:

• “There’s no point in sharpening the ax when the chainsaw is coming out tomorrow.” (49:20–49:36, Scott)

On what not to automate:

• “The minute I see [AI-generated writing tropes]…if it wasn't worth the time for this person to write this, then it's not worth the time for me to read it.” (59:24–59:58, Scott)

On legal AI vs. code AI:

• “AI for code and AI for legal…I think in both of these areas, you're not trying to write creative original code or creative original contracts. You're trying to make functional documents.” (61:08–61:26, Scott)

On Keith Rabois’ communication style:

• “He has five bullet points about why this opportunity is incredible...communicated everything in like five minutes...then he was like, okay, I think we're done.” (89:56–90:13, Scott)

Timestamps for Noteworthy Segments

• Spellbook’s market position and product overview: 00:06–04:38
• Why a Word plugin, technical hurdles: 10:20–12:12
• Legal market AI transformation: 13:49–19:21
• Aggregated data product and privacy: 23:49–24:49
• On RAG vs. fine-tuning models: 26:32–30:25
• Bottoms-up sales strategy: 37:33–42:26
• Legal AI landscape, consolidation: 42:26–46:56
• Rapid adaptation & future of agentive AI: 47:22–56:59
• Spellbook’s early journey, lessons from landing pages: 66:48–74:40
• Fundraising, working with Keith Rabois: 83:03–90:41
• Scott’s personal AI stack: 91:00–94:35

Summary

This episode is essential listening for anyone curious about how generative AI is reshaping “traditional” professions, especially law. Scott Stevenson’s story offers invaluable lessons in product development, grit, and the necessity of adapting—quickly!—as the ground shifts beneath your feet. Spellbook’s rise offers a playbook for building workflow-embedded, industry-specific AI products that deliver unique value far beyond what generalized tools can provide.