AI Agents Are Only As Good As The Tools They Can Reach

Every serious software company now supports Model Context Protocol (MCP), which has transitioned from experimental to standard in under two years. According to a December 2025 Stacklok survey, nearly half of technical leaders in software reported using MCP in production. Best-in-class MCP is now expected, and the critical question is what your agents can access through it—particularly real customers and their insights.

While AI agents have dramatically accelerated building products, they haven't similarly sped up learning from customers. The teams that gain a competitive edge are those that integrate customer feedback into every stage of the AI product development lifecycle (AI-PDLC), rather than treating it as a separate final step.

The Accelerated Development Dilemma

For 20 years, product development has continuously accelerated. The AI-native PDLC represents the latest leap, where agents draft specifications, generate prototypes, write code, and run quality assurance. While the stages remain similar, the pace has transformed—work that once required a sprint can now be completed in an afternoon.

Research, validation, and feedback cycles have similarly shortened from weeks to hours or days. Human feedback must remain rigorous but shouldn't impede development. The goal is to learn at the same pace you build, maintaining speed without losing sight of the product's purpose and value.

When AI output can be generated in hours, end-of-lifecycle customer feedback studies that take too long can negate any speed advantages gained. Yet these studies remain essential because AI output often appears finished but is rarely complete and requires validation.

Research Becomes Accessible

Working with Zigzag, a puppy-coaching app, Auros helped develop and refine its AI chat feature Ziggy. The team conducted numerous tests throughout development using real-time customer feedback to determine feature readiness. In the first six months, Ziggy achieved a 96% positive customer rating, was used over 80,000 times, and reduced human trainer support requests by 27% thanks to continuous, high-quality user input.

Customer feedback delivers maximum value when it reaches product teams quickly enough to enable iteration rather than requiring product rebuilds. This is particularly crucial when agents and AI are involved in development, as end-user feedback will determine success and must arrive promptly.

Balancing Speed And Insight

The temptation to use synthetic users for feedback is strong, but companies should proceed cautiously. Models are designed to produce the most likely answers, so having one model validate another's output may confirm typical patterns while missing surprising insights that often contain the most product value.

Teams should map where in their development process they want agents to reach real customers. Most find agents handle build steps completely but lack routes to customer input until a human initiates a study. This gap needs addressing by giving agents early access to human insights.

Agent access to people should be treated as a governance decision rather than a simple feature switch. Companies need rules for spending (to control recruitment costs) and approval processes (requiring human sign-off before agents launch anything). Product leadership should consider three key questions: at what lifecycle stage can agents ask customers questions, who decides when this is allowed, and are we measuring build speed or learning speed?

The Competitive Advantage

MCP support became standard within two years, and agent-driven build speed is following the same trajectory. Once all teams can build in hours, the advantage will belong to organizations that incorporate real customer insight into their AI-PDLC and retrieve it quickly. Agents are helping to shorten the distance between decision-making and gathering evidence from affected users.

Companies should identify which decisions they want agents to make, provide appropriate guardrails, connect research results to the build process, and measure the speed of learning—all while maintaining human involvement at every level. The teams that embed research throughout the AI-PDLC and keep real people in the loop as agents take on more work will set the new standard for product development.