The sensible defaults gap: what most top SaaS products get wrong for AI agents

AI agents are attempting to operate software at scale, but they're encountering a fundamental problem: most products aren't designed with reasonable starting states. According to a recent audit of the top 100 SaaS products, sensible defaults—one of eight Agent Factors Engineering principles that measure agentability—averages just 35 out of 100.

This gap matters because AI agents, unlike human users, cannot easily intuit what missing configuration values should be or recover gracefully from ambiguous initial states. When product teams haven't considered sensible defaults as part of their agent operability strategy, they create friction that compounds across every automated workflow.

What agentability measures

Agentability is a score from 0 to 100 that quantifies how well AI agents can operate software. The Agentability Project evaluates products across eight Agent Factors Engineering (AFE) principles:

• Machine readability: structured data and semantic markup
• Transparency: observable system state and rules
• Shadow-UI avoidance: preference for standard elements over custom implementations
• Sensible defaults: reasonable starting values that minimize required decisions
• Control: configuration options and operational boundaries
• Chunking: decomposable workflows and modular operations
• Status: clear feedback on operation outcomes
• Clean handoffs: defined transition points between automated and human work

In the audit of 100 products, the average agentability score was 38.3 out of 100. The distribution shows 22 products classified as Agent-Ready (scoring 45 or above), 54 as Developing (35-44), 17 as Lagging (20-34), and 7 as Agent-Blind (under 20).

Why sensible defaults matter for agents

Sensible defaults reduce the decision space an AI agent must navigate to complete a task. When a form requires 15 fields but only 3 are truly essential, and the other 12 have reasonable defaults, an agent can proceed with minimal input. When those 12 fields are empty and required, the agent must either guess, prompt a human for values, or fail the operation entirely.

Human users tolerate poor defaults because they can draw on context, organizational knowledge, and intuition. An experienced user sees an empty dropdown for "Report Format" and knows that PDF is the standard choice for their use case. An AI agent sees an enum with eight possible values and no signal about which to select.

This problem manifests in several patterns observed across the audited products:

• Required fields with no pre-populated values, forcing agents to make arbitrary choices
• Configuration wizards that demand upfront decisions about options that could have sensible starting states
• Settings pages with dozens of toggles, all in an undefined or neutral state
• Workflows that require explicit selection from dropdown menus even when 80% of users choose the same option

The broader AFE context

Sensible defaults doesn't exist in isolation. The audit revealed significant challenges across multiple AFE principles. Transparency averaged just 5 out of 100, with 83 of 100 products scoring zero. Shadow-UI avoidance averaged 17, with 80 of 100 products scoring under 20. These low scores indicate that most products prioritize visual design and human-facing interfaces over structured, agent-readable foundations.

By contrast, machine readability averaged 71—the highest of any principle. This suggests that basic semantic HTML and structured markup are relatively common, even if products haven't considered agent operability holistically. Control (48), chunking (45), status (47), and clean handoffs (40) all clustered in the mid-range, indicating partial consideration but significant room for improvement.

The sensible defaults score of 35 sits roughly in the middle of this spectrum, which makes it a particularly actionable principle. Unlike transparency or shadow-UI avoidance, which may require architectural changes, improving defaults is often a matter of product discipline and user research translated into configuration values.

What this means for product teams

Product teams building for agentability should audit their own sensible defaults implementation. The Agentability Project offers tools to run a free audit and compare results against the Agentability Index.

Concrete improvements

Start by identifying every decision point in core workflows. For each field, toggle, or selection:

1. Determine what percentage of users choose each option based on actual usage data
2. Set the most common choice as the default value
3. For fields where no clear majority exists, consider whether the field is truly necessary or could be optional
4. Document the reasoning behind each default so future product changes don't inadvertently break agent workflows

Pay special attention to onboarding flows and account creation. These are often the first places agents interact with your product, and poor defaults here create compound friction. If 90% of programmatic account creation follows the same pattern, make that pattern the default path.

Testing with agents

Validate defaults by attempting to complete workflows with minimal input. Can a task be accomplished by providing only the truly variable information, with everything else falling back to reasonable values? If an agent (or a script, or an API call) can't complete a workflow without specifying every parameter, your defaults likely need work.

Balancing defaults with control

Sensible defaults should not eliminate optionality. The control principle (averaging 48 in the audit) measures whether agents can override defaults when needed. The goal is to make the common case trivial while preserving the ability to handle edge cases. Provide defaults, expose overrides, and document both clearly.

Moving forward

The 35-point average for sensible defaults across top SaaS products represents a clear opportunity. Unlike some AFE principles that require fundamental architectural changes, defaults can be improved through systematic product work: research, configuration, and testing. As AI agents become more prevalent in software operations, products that invest in sensible defaults now will have a significant advantage in agent-mediated workflows.

Product teams don't need to solve all eight AFE principles simultaneously. Sensible defaults offers a practical starting point with measurable impact and relatively contained scope. The products that address this gap will find their software more operable not just by agents, but by API consumers, automation tools, and power users who benefit from the same reduction in decision overhead.