The 3x3x3 Framework: From Vision to Working Product

AI has made building dramatically faster. It has not made deciding what to build any easier.

For years, product teams worked under a simple constraint: software was expensive to create. Turning an idea into something customers could use required designers, engineers, infrastructure, planning, and multiple development cycles. Because building was slow, teams spent weeks preparing prototypes, specifications, and experiments.

That constraint is disappearing. Today, a founder, product manager, designer, or domain expert can describe an idea in natural language and generate a working interface, database, authentication flow, and basic product logic in hours. AI app builders can move from a prompt to a deployed prototype, while coding agents can write tests, debug problems, document code, and implement complete workflows.

This changes how products can be built. But it does not change the fundamental challenge of product development: building something is not the same as discovering something worth building. In fact, when software becomes cheaper to produce, the risk of building the wrong thing increases. Teams can now create polished products before they have properly understood the customer, problem, behavior, or business model.

The bottleneck has moved. It used to be implementation. Now it is judgment.

The AI-era 3x3x3 framework

To operate effectively in this environment, we need a framework that connects long-term vision with short-term learning, while taking advantage of the speed AI gives us. My adaptation of Ash Maurya's original 3x3x3 perspective is designed for exactly that.

The original framework connects three time horizons: a three-year vision, a three-month strategy, and a three-week experimentation cycle. The purpose is to keep vision, strategy, and product decisions aligned instead of allowing teams to rush directly into building features.

In the AI era, the first two horizons remain valuable. The third one changes. The updated model is:

3 years for direction · 3 months for strategic progress · 3 days for product evidence

The three-year horizon tells us where we are going. The three-month horizon tells us which major uncertainty we are trying to remove. The three-day horizon turns an assumption into something customers can see, use, and react to. This is not simply a faster development process. It is a faster learning system.

1. Three years: Define the change you want to create

AI makes it tempting to begin with a prompt: 'Build me an AI assistant for sales teams.' 'Create a platform that helps managers coach employees.' 'Make an app that automates financial analysis.' These prompts may produce working products, but they are not visions.

A product vision should describe the meaningful change you want to create for a specific group of people. It should remain useful even as interfaces, models, features, and technologies change. Start with two questions. Why should this product exist? What meaningful problem are you trying to solve? Who experiences this problem? Why is solving it important? What should become easier, faster, safer, more accessible, or more effective because your product exists?

A weak answer describes the technology: 'We want to build an AI-powered project management platform.' A stronger answer describes the change: 'We want small teams to coordinate complex work without spending hours maintaining project plans and status reports.' The first statement can become outdated as technology changes. The second gives the team a durable direction.

A vision needs a measurable outcome. This is not a detailed financial forecast. It is a minimum success criterion that makes the ambition concrete. For example: 'Within three years, 10,000 small teams use the product every week to plan work and communicate progress.' Or: 'Within three years, customers complete their monthly financial reporting in one day instead of one week.'

The combination of purpose and measurable impact creates a decision filter. When a new feature, model, trend, or opportunity appears, you can ask: does this help us create the three-year change we committed to? Without this filter, AI speed creates strategic noise. Every new capability looks like a reason to change direction. A clear vision protects the team from becoming a collection of demos.

2. Three months: Choose the uncertainty that matters most

A three-year vision is too broad to guide daily work. The next horizon is a three-month strategic outcome. In the original 3x3x3 model, strategies are timeboxed so that teams can decide whether to persevere, pivot, or reset. This principle becomes even more important when AI allows teams to pursue many ideas at once.

The question for the next three months is not: which features should we build? It is: which uncertainty must we remove before we can make the next major decision? Most early products face uncertainty in one or more of these areas:

Problem uncertainty. Is this problem important enough? How frequently does it occur? Are people already spending time or money trying to solve it? · Solution uncertainty. Does our approach create a meaningfully better outcome? Can customers understand the product without a long explanation? Can they reach the expected value? · Behavior uncertainty. Will customers change their existing habits? Will they return? Will they trust AI to perform this task? Where do they require control, review, or approval? · Business uncertainty. Who pays? How much is the outcome worth? Can we reach customers efficiently? Can the product be delivered with sustainable margins? · Technology and risk uncertainty. Can the system produce reliable enough results? What happens when the AI is wrong? Which data can safely be processed? Which actions require human approval?

Choose one primary uncertainty for the quarter. For example: 'Over the next three months, we need to learn whether independent consultants will trust an AI system to prepare the first version of a client proposal.' The goal should include evidence: 'We will consider the strategy promising if 20 consultants use the product for real proposals, at least 10 use it more than once, and five agree to pay for continued access.' This creates a strategic boundary. The team is not trying to build a complete proposal management platform. It is trying to test whether a specific customer will delegate a specific job to AI. That distinction can save months of unnecessary development.

3. Three days: Turn assumptions into product evidence

This is where the AI-era version differs most from the original model. A three-week experimentation sprint made sense when prototypes required substantial design and engineering effort. Today, a small team can often create a functional test within a few days. Tools can generate interfaces, databases, integrations, application logic, and deployable environments through natural-language instructions. Coding agents can also assist with implementation, testing, debugging, and documentation.

The product experimentation cycle can therefore be compressed from three weeks to three days. But only when the objective is carefully constrained. A three-day sprint should test one important assumption. It should not attempt to create a complete product.

Day 1: Frame the decision

Before opening an AI builder, write down the assumption, the riskiest unknown, the evidence needed, and the decision you will make. For example: 'We believe consultants will upload previous proposals because the generated output becomes significantly more relevant. The riskiest unknown is that consultants may consider previous proposals confidential and refuse to upload them. We need at least five of eight consultants to upload a real document without being encouraged to do so. If they upload and report improvement, we will invest in secure document ingestion. If they refuse, we will explore structured onboarding without document uploads.'

This step prevents the team from measuring activity instead of learning. 'Users liked the prototype' is not strong evidence. 'They uploaded confidential work, generated a proposal, edited it, and asked to use it for another client' is much stronger.

Day 2: Build the smallest credible experience

Now use AI to create only what is required to test the assumption. This may include a working interface, a simulated AI response, a basic database, a document upload, a payment page, a human-operated workflow behind the interface, or one complete end-to-end customer journey. The prototype should be credible enough for the customer to behave realistically. It does not need to be scalable, automated, or architecturally complete.

This is where the distinction between a prototype and a production system matters. Vibe coding is excellent for exploration, but code intended for real production use still requires clear specifications, testing, security controls, boundaries, and human review. Use AI speed to reduce the cost of learning, not to avoid engineering discipline.

Day 3: Put it in front of customers

The third day is not another building day. It is a customer evidence day. Ask customers to complete a realistic task. Avoid explaining every part of the experience. Observe where they hesitate, what they expect, and whether they can reach the outcome independently.

Did they provide real data? · Did they complete the workflow? · Did they correct the AI output? · Did they save, export, or share the result? · Did they return with another task? · Did they invite someone else? · Did they ask how much it costs? · Would they be disappointed if they could no longer use it?

At the end of the day, make one of three decisions: continue because the evidence supports the assumption; change because the problem is relevant but the workflow, audience, positioning, or solution needs adjustment; or stop because the assumption is not supported and further investment is not currently justified. A sprint is complete when it produces a decision, not when it produces software.

The three layers of evidence

AI-generated prototypes can look highly convincing. This makes it easy to confuse visual quality with validation. To avoid this, evaluate every three-day sprint through three layers of evidence.

Conversation evidence. What customers say. This is useful for understanding language, context, current processes, and perceived pain. However, customers often say they would use a product without changing their actual behavior. · Behavior evidence. What customers do. Do they upload data, complete a task, return, share the output, connect an account, or invite a colleague? Behavior is stronger than positive feedback. · Commitment evidence. What customers are willing to risk. Will they pay, sign a letter of intent, provide sensitive data, introduce a colleague, schedule another session, or replace an existing tool? Commitment is the strongest early signal because it carries a cost.

Every sprint should aim to move from conversation toward behavior and commitment.

AI changes the prototype, not the scientific method

The classic product discovery loop still applies: identify an assumption, design an experiment, build what the experiment requires, observe customer behavior, and update the strategy. AI changes the cost and speed of the build step. It does not remove the need for the other steps.

This is a critical distinction. A team that uses AI only to generate more features may move faster while learning very little. A team that uses AI to run more focused experiments can create a genuine competitive advantage. The goal is not maximum output. The goal is maximum learning per unit of time.

Example: Building an AI meeting assistant

Imagine that we want to create an AI product that helps product managers turn customer conversations into actionable product insights. The three-year direction is: help product teams understand customers without losing important evidence across recordings, documents, and internal conversations. The minimum success criterion: within three years, 2,000 product teams use the product weekly, and customers report that it reduces research synthesis from several days to a few hours.

The three-month strategy: the primary uncertainty is whether product managers will trust AI-generated themes enough to use them when prioritizing product decisions. The quarterly evidence: 30 product managers upload real customer interviews, 15 use the product repeatedly, and at least five reference the generated evidence in a roadmap or product review.

The first three-day sprint tests the assumption that product managers will find evidence more trustworthy when every AI-generated insight is connected to exact customer statements. Day 1 defines the workflow and success criteria. Day 2 builds a prototype where users upload interview transcripts, receive several themes, and open the original supporting statements for each theme. Day 3 tests the experience with five product managers using their own research. The team observes whether they inspect the evidence, correct the themes, export the results, and use the output in a real decision.

The team does not need collaboration, permissions, billing, integrations, automated transcription, or a complete research repository yet. Those features may eventually matter. They do not matter for the current decision.

From product manager to product builder

This model also changes the role of product people. Product managers were traditionally responsible for identifying problems, defining requirements, aligning teams, and deciding what should be built. Designers created the experience, while engineers implemented it. Those boundaries are becoming more fluid.

Product people can now create working interfaces, connect services, generate application logic, and test complete workflows without waiting for a full delivery team. This does not eliminate the need for designers and engineers. It changes when and how their expertise is applied.

Instead of using engineering capacity to test every early assumption, product people can arrive with better-defined problems, working prototypes, real customer behavior, identified technical risks, clearer requirements, and evidence that the opportunity deserves investment. Engineers can then focus on architecture, security, reliability, performance, maintainability, and difficult technical problems. Designers can focus on complex interactions, systems, trust, accessibility, and experience quality. Product managers become more capable of translating customer insight directly into testable software. The best product people will not only write requirements. They will build evidence.

The danger of infinite building

There is a new product failure mode emerging. Because prototypes are inexpensive, teams keep creating them. One version becomes five. One feature becomes ten. Every customer request can be implemented immediately. The result is activity without strategic progress.

The solution is not to slow down AI-assisted building. The solution is to create stronger constraints. For every three-day sprint, define one assumption, one customer segment, one important workflow, one primary metric, and one decision. Anything that does not contribute to that decision should wait. AI gives us almost unlimited implementation possibilities. Product strategy is the discipline of deciding which possibilities do not matter yet.

A practical 3x3x3 canvas

Before starting a new product or major initiative, complete the following and keep it on one page. If the team cannot express the experiment clearly, it is probably trying to test too many things at once.

Three years. Purpose: what meaningful change should exist? Customer: for whom? Minimum success: what measurable outcome would make the journey worthwhile? · Three months. Primary uncertainty: what must we learn next? Strategic approach: how do we intend to learn it? Success evidence: which behaviors or outcomes would justify continuing? · Three days. Assumption: what do we believe? Prototype: what is the smallest credible experience that can test it? Decision: what will we continue, change, or stop based on the evidence?

Build faster, but decide more carefully

AI has reduced the distance between an idea and working software. That is a significant advantage, but only when paired with discipline. The winning teams will not necessarily be those that generate the most code or release the most features. They will be the teams that connect three levels of thinking: a clear long-term direction, a focused strategic uncertainty, and a rapid product experiment.

Three years provide direction. Three months provide focus. Three days provide evidence. The result is a product organization that can move quickly without becoming reactive, build continuously without becoming feature-driven, and use AI without confusing generated software with validated customer value.

In the AI era, almost anyone can build a product. The harder and more valuable skill is knowing what deserves to be built next.