Building AI products competitors can't match

How do you win when everyone has access to the same AI models?

On September 10th,

Reforge

held its first AI Product Summit.

Diana Stepner

wrote a fantastic recap, highlighting one of my favorite talks, Why Humanities Majors Are the Secret to Product Success by Shir Yehoshua, Head of AI at Notion.

Over the last couple of years, I’ve spent a lot of time with companies helping them navigate the AI shift—a time that is opportunity rich, but fraught with challenges for companies big and small.

Brian Balfour

and I built the Reforge AI Strategy program to help leaders win in the most intense environment we’ve seen in the history of tech. The course is available on demand and our second cohort starts on October 16th.

We have a fantastic line-up of guests including:

• Justin Farris, VP of Product at Read.ai

• Laura Burkhauser, CEO at Descript

• Darius Contractor, Chief Growth Officer at Otter.ai

• Jeff Chow, Chief Product & Technology Offier at Miro

As a sneak peek, here is just one of many concepts we’ll cover in the AI Strategy program: how to build AI products competitors can’t match.

In the book Blue Ocean Strategy, the authors describe two types of markets:

• Blue Oceans are untapped markets where competition is limited and opportunities are rich.

• Red Oceans are crowded markets where established players fight intensely for market share.

In Red Ocean markets, companies have to beat the competition and take market share. As a result, companies often compete on price and margins suffer due to higher marketing costs and lower prices.

In Blue Ocean markets, competition is irrelevant or obsolete. Companies succeed by creating market share, benefiting from innovative products, healthy profits, and strong defensibility.

Over the last 50 years, tech has unfolded in three waves: the PC wave, the Internet wave, and the Mobile wave. Historically, the transition between waves (the “platform shifts”) have created blue ocean opportunities. New startups have moved quickly to capitalize on the new technology, while incumbents struggled to orient themselves around the new reality.

Why? Because the PC, Internet, and Mobile waves disrupted incumbents, undermining the advantages they had established. For example, Adobe needed to rethink and rebuild its design products for cloud-based collaboration, opening up a window of opportunity for Figma to build an Internet native product.

This disruption created fertile ground for new companies to emerge as winners. Each wave has seeded pillar companies, leading to an unprecedented rally of wealth creation:

• Personal Computing → Microsoft, Intel, Oracle, Adobe

• Internet → Google, Facebook, Salesforce, Figma

• Mobile → Apple, Uber, TikTok

This time is different. In many cases, AI is accelerating, rather than disrupting incumbents.

Let’s go back to our Adobe example. While Adobe wasn’t able to beat Figma to the cloud collaboration market, they have moved quickly to layer generative AI across their portfolio. This includes adding AI features into existing products, like Generative Fill and Generative Expand, and launching new standalone products like Adobe Firefly. Adobe is still far from parity with Figma’s collaboration features, but they have quickly leapfrogged on AI features—closing off AI opportunities for hopeful startups. (Check out Will AI create massive opportunities for startups? for a deeper dive into disruptive vs. accelerating technologies).

Today, AI products find themselves playing in a competitive, lightning fast environment. Competition is coming from every side:

• fast-moving AI platforms like OpenAI and Anthropic

• established tech companies like Adobe and Microsoft

• and a raft of well-funded startups

So one question needs to be at the top of every product leader’s mind:

How do we win in the most intense environment in the history of tech?

Let’s look at how you can use AI, including off-the-shelf AI models, to create an unfair advantage for your product and your company.

Many companies start by trying to reinvent the wheel. They assume AI advantage comes from advancing the state-of-the-art, and teams spend months or years developing custom models & infrastructure.

Certainly, that is a viable path—if you can lead the pack. AI labs, like OpenAI and Anthropic, have created massive value by innovating at the model layer,

But, this approach isn’t feasible for most companies. The cost of ML talent and training is prohibitive. And sometimes even that isn’t enough. Inflection, for example, had all the cards, but faced an uphill battle competing with the growing landscape of roughly equivalent models.

For most companies, its not feasible to advance the state-of-the art with AI… but its also not necessary. Over the last few years, a remarkable thing has happened. The most advanced AI capabilities are not being hoarded by the biggest tech companies. Instead, AI is now democratized—available via easy-to-use, affordable APIs and open source models.

You don’t need to reinvent the wheel, but you can’t get away with underinvesting. Some companies make the mistake of sprinkling basic AI features, like chatbots, onto their products.

While these features are easy to add, they rarely deliver meaningful value to users and can be quickly replicated by competitors.

Instead of being on one end of either of these extremes, there is a third approach: thinking of AI capabilities as building blocks that can be assembled in unique combinations with your product and its data… just as LEGO bricks can create endless possibilities.

The most effective AI features are built from three elements that, when combined, create competitive advantage:

1. AI Capabilities

2. Your Data

3. Your Functionality

AI products start, of course, with AI. Powerful AI models enable us to build products that would have seemed magical less than 2 years ago. Many of those models are just an API call away.

So if everyone has access to the same AI capabilities, then how do we create AI-powered products that stand out?

Your competitive advantage… your unfair advantage… comes from what is uniquely yours:

1. Your data

2. Your product’s functionality

3. Your understanding of unmet customer needs

Let’s look at these in more detail.

Your data provides context for AI models—it is the most important element to turn off-the-shelf AI capabilities into something bespoke for your product.

Next, we need to think about what your product does (its functionality). Your product’s functionality determines how AI behaves—it gives AI superpowers that can’t be had anywhere else.

Every AI feature is assembled from these three elements—the AI capabilities you choose, your data, and your functionality.

Next, we need to determine how the three elements connect to each other. These interactions are where true differentiation happens.

You data provides context to the AI—giving it the necessary information for AI to perform its task effectively. Context is critical to turning off-the-shelf AI models into custom solutions for your user’s needs.

For example, a medical diagnostic tool can feed patient history into an AI to help identify potential conditions.

AI generates output such as insights, content, or predictions. This output creates value in two ways. First, by providing results your customers want and second by becoming part of your proprietary data set.

For example, a content management system might help people use AI to tag and categorize articles, helping customers organize their content while enhancing your proprietary metadata.

Orchestration is the next piece of the puzzle. Your product determines when to use the AI—whether directed by the user or triggered automatically in the background. Importantly, AI does not replace traditional software. Instead, traditional software and AI complement each other, each excelling in different situations. Your product orchestrates the interactions between these different types of systems.

For example, an email client might offer AI writing suggestions when it detects the user pausing for more than 5 seconds.

Finally, let’s look at one of the most exciting developments. AI is getting increasingly good at using tools. AI can learn when to use your product, giving it unique superpowers.

For example, a customer support AI agent can be configured to use a payment system as one of its tools. Then, the agent can decide to automatically initiative a refund when it identifies eligible use cases.

We are at the very beginning of what’s possible with tool use, and this will be one of the most exciting areas as agents can increasingly autonomous. (Check out Tool Augmented Generation for a deeper dive.)

The power of this approach lies in creating combinations that competitors cannot easily replicate.

While they may have access to the same AI models, they won’t have your unique data and functionality—making your AI features distinctively valuable and difficult to match.

AI doesn’t just create new capabilities—AI magnifies the value of what your company already has.

Let’s look at an example…

Over the last 18 months, Miro has evolved into an AI-powered collaboration space that helps teams supercharge their teamwork. Underlying their strategy is a simple, powerful principle:

The canvas is the prompt.

This approach has led to a portfolio of AI features, with AI prototyping standing out as one of the most powerful. Teams can transform a pile of messy notes into a clickable prototype in moments—all without leaving the board.

Importantly, Miro is using off-the-shelf AI models to build features others can’t easily replicate. They do this by integrating their data and product functionality deeply into the AI experience. Here’s how the blocks connect:

• Context: The entire board’s content provides situational awareness to the AI

• Output: The AI generates a prototype based on the current work, which becomes part of the canvas

• Orchestration: Prototypes can be invoked when users need and the system asks questions as needed for additional context

• Tool Use: The AI uses Miro’s canvas as a tool, placing the prototype on the board. Another Miro feature, AI Sidekicks, adds comments from an “Agile Coach” or “Product Leader” just like any other user would.

Miro is using the same AI capabilities available to everyone, but to power a feature that only they can deliver.

Let’s look at how you can use this approach for your own products.

Every day, new AI capabilities emerge. Just this year, we’ve seen massive improvements in deep reasoning and video generation — things that seemed like science fiction a few years ago.

With all these AI capability, we have more ways to solve customer problems than every before. But, we need to be wary of solutions in search of problems.

Just because AI can do something, doesn’t mean it should.

Always start with the customer and work backwards to identify which AI capabilities can solve their unmet needs.

The key to powerful AI features lies at the intersection of what’s possible and what users want.

Descript has reimagined video editing: instead of working with cumbersome timelines, creators edit videos through AI-transcribed scripts. Although AI is core to how Descript works, they don’t emphasize the technology. Instead, they package advanced AI features around a very human need: helping creators look and sound their best.

Take their Eye Contact feature, which seamlessly edits videos so creators appear to maintain constant eye contact with the camera rather than looking at their notes. Another feature, Studio Sound, uses AI to enhance audio quality, delivering professional results regardless of where creators record.

The most exciting thing about being a product builder today is that we have an ever-growing menu of AI building blocks to choose from. We can translate, edit, manipulate, understand, and generate text. The same capabilities exist for images, video, and audio. We have increasingly powerful reasoning models that can plan tasks, use tools, and coordinate with other agents.

With this vast toolkit available, the key question becomes: which unmet customer needs can now be solved? This is where selecting AI capabilities should start—not with what’s shiny and new, but with identifying important customer problems that have shifted from impossible to achievable.

Once you’ve identified which AI capabilities can address real customer needs, you’re ready to determine how those capabilities integrate with your existing data and functionality.

As we’ve discussed, your data is crucial to building effective AI features—its provides the essential context that AI models need to deliver relevant, personalized results.

Without proper context, even the most sophisticated AI models produce generic outputs. Your proprietary data represents a key competitive advantage because it contains unique information that your competitors don’t have access to.

You can tap into various types of high-value data:

• Real-time data that provides context LLMs won’t have in their training datasets (flight prices, inventory availability, traffic conditions)

• User-specific data that is highly personalized and which competitors can’t access (listening or viewing patterns, healthcare records, purchase history)

• Domain-specific data that is high-value and hard-to-find (legal cases, medical research, editorial content)

• Human judgment data that augments existing sources with curated content (curated images, ratings & reviews, social proof)

• Reinforcement data that helps you improve results over time (thumbs up/down on responses, user corrections)

The most powerful AI features often emerge when you connect multiple types of data—like combining historical usage patterns with real-time user context to provide predictive assistance.

Your product’s functionality encompasses everything your product can do—both what’s visible to users and what happens behind the scenes.

The relationship between AI and your product’s functionality works both ways:

• Your product determines when to use AI to solve customer problems.

• But, increasingly, AI can use your product like a tool—tapping into your product’s functionality as part of generating output.

You can tap into various types of high-value functionality:

• Specialized workflows, such as contract review pipelines, financial approval processes, or medical prescription management

• Unique algorithms, such as recommendation or personalization systems

• Business rules that encode decision criteria or implement complex requirements (insurance underwriting rules, loan approval criteria)

• Integrations into multiple systems, such as integrating with a healthcare records system or payment system

Let’s look at another example…

Granola is one of the fastest growing AI apps. The just raised another $43 million to accelerate their vision of AI-powered note-taking.

But, AI note-taking is a Red Ocean, filled with competitors like Otter, Fireflies, Fathom, and integrated AI note-taking in Zoom, Teams, Notion, Google Meet, and ChatGPT.

How did Granola break through?

They identified an unmet customer problem.

Most of the existing note taking products take over the note-taking process… they act as a secretary that people can delegate to.

However, for an important segment of users, note taking is a critical part of their job. They don’t want someone else to take their notes… they want to take better notes.

So, Granola created a platform for them… for the person that wants AI to amplify rather than replace.

Note: This spectrum from autonomy (I want AI to do the job) to amplification (I want AI to help me do the job better) is an important framework for understanding product positioning. In nearly every category, we have AI products at varying points along the spectrum. For example, AI coding tools like Lovable are for people who want the AI to write the software for them, while tools like Claude Code and Cursor are for engineers who want AI to amplify their skills.

Granola is one of the fastest-growing AI companies, yet they rely entirely on off-the-shelf AI systems.

For transcription, they use Deepgram. For LLM features—note generation, the ability to chat with notes, and generating artifacts like follow-up emails—they rely on Anthropic and OpenAI.

Despite relying on off-the-shelf AI systems, Granola has assembled these capabilities into something uniquely powerful. Here’s how:

• Context - Their context includes both the notes users take and the transcriptions they generate

• Output - They generate enhanced notes that become part of an ever-expanding repository of knowledge—a second brain

• Orchestration - Granola is implemented as a Mac app, not a web app, giving it the ability to detect when meetings start, access system-level audio, and automatically trigger transcription

• Tool use - Calendar integrations provide metadata about meetings, including attendees and topics

They have assembled a unique, differentiated AI product—built not by developing proprietary AI models, but by understanding an unmet customer need and crafting a solution that elegantly addresses it.

I’ve seen a lot of the history of tech first-hand. This is the most intense time I’ve ever seen—ripe with opportunity, but fraught with challenges.

Winning the AI era is hard, but not impossible. It won’t require hard-to-find ML researchers and hundreds of millions of dollars to train frontier models. It will require a deep understanding of customer problems, emerging AI solutions to those problems, and the unique role your product can play.

To get started, ask yourself four questions:

1. What unmet customer problems need to be solved?

2. What AI capabilities can solve those problems in novel ways?

3. What proprietary data can power those solutions?

4. What superpowers can our product give to AI?

Winning with AI isn’t just about the technology—it’s about turning what’s uniquely yours into a competitive moat.

This is one of many powerful concepts we cover in the AI Strategy course, designed to give you the ideas, frameworks, and tools to navigate today’s unprecedented opportunity. We built the Reforge AI Strategy program to help leaders win in a rapidly evolving landscape. The course is available on-demand and our second cohort starts on October 16th.