Claim
When AI agents take over code generation, what a product team ships is no longer the code itself. The product becomes the context, judgment, and intent that directed what was built. Customers buy outcomes and decisions, not output.
Mechanism
Code has always been an artifact, not the value. The value is solving the right problem well. When code generation is cheap and fast, the scarce input shifts from production to direction. A team that frames the problem correctly and gives the AI accurate context produces better outcomes than the team that writes more or faster code.
Saarinen names this as a risk hidden in the AI speed gain: teams can ship more software while learning less about their customers. The constraint moves from execution to understanding. The product leader's job shifts from managing output to managing the quality of context: does the team know enough about the user's actual problem to direct AI toward the right solution?
Conditions
Holds when: code generation is fast enough that speed of development is no longer the competitive bottleneck, and multiple teams can build functionally similar things with similar effort.
Fails when: the technical problem is still primarily a code-difficulty problem, or when first-mover speed creates durable competitive advantage independent of solution quality.
Evidence
From Saarinen's talk at the Lenny and Friends Summit 2026:
"The output is not the product. Customers are not buying the lines of code"
Saarinen frames this as the central tension: teams that accelerate development without increasing customer-learning cycles are optimizing for a commodity. The teams that invest in understanding customers and problems, not just building faster, will differentiate on the one input that remains scarce.
Signals
- Teams using AI to accelerate development but not increasing customer-facing research cycles are optimizing for the commodity side of the stack.
- When most teams can build the same thing in the same time, the differentiating variable becomes clarity of context about the problem, not speed of execution.
- Product leaders who still measure primarily by feature velocity are measuring the part of the job AI has made less valuable.
Counter-evidence
The argument assumes code generation is effectively solved. For many product categories, deep technical differentiation still matters. Infrastructure, performance-sensitive systems, and novel algorithm development remain domains where the code itself carries value beyond the intent behind it. Context quality also depends on having done the customer learning in the first place; AI does not supply that work.
Cross-references
- ins_saarinen-context-is-the-product is the first card from Karri Saarinen in this corpus.