More AI Agents mean more Agent Slop. Here's how to fight it with Wayfound.
OpenAI and Anthropic recently announced significant enhancements to their Agent SDKs and builders – further accelerating the already rapid pace of agentic experimentation and adoption. It is now easier than ever to create true multi-agent systems, but it’s also easier than ever to create agent slop. MIT found in summer 2025 that 95% of generative AI prototypes fail, around the same time that Accenture reported only 8% of $1 billion-plus organizations have scaled-up AI via multiple initiatives.
As this new architecture evolves, it is quite clear that agents need supervisors to verify and guide their work, otherwise the ROI just won’t materialize. Pilots and prototypes will be abandoned, but worse is that agents moved into live, production settings will create more (not less) burden and possibly even increase business risk.
Wayfound fills in the critical gap that most agent building approaches consider as an afterthought: Making sure that agents perform successfully for the business. To do this, Wayfound has created the first no-code agent supervisor platform for this exact purpose of monitoring and improving agents against defined goals, ensuring compliance, and reporting on the impact. The bottom line: Wayfound makes sure more agents find their way to ROI.
How Wayfound Works with Platforms like OpenAI and Anthropic
A key feature of those agent builders is their native support for Model Context Protocol (MCP). The key advantage of using MCP with agents is that you don’t have to program a discreet workflow. The AI decides what MCP tools to use and when, guided by simple prompts and tool descriptions. Wayfound’s MCP server provides several important tools, one of which is evaluate_session. By giving your agent access to the Wayfound supervisor, it can verify its work against a set of guidelines that have been curated by the product or business owner of the agent. No change to the codebase or agent configuration is required as guidelines evolve and change.
Case Example Using Wayfound
To demonstrate this architecture, we built a multi-agent stock researcher application using the Anthropic Agent SDK to research stocks and write reports. Given a stock symbol, the AI Coordinator agent leverages two subagents, web-researcher and report-writer, and a Wayfound supervisor agent via MCP. When the report-writer subagent finishes writing its report, Anthropic’s orchestration agent understands that it needs to verify the work using the Wayfound supervisor.
- The generated report is checked against the guidelines defined in Wayfound
- The results of the check are returned to the coordinator agent
- The coordinator agent decides whether the report meets expectations or needs to be re-written, based on the analysis and recommendation from Wayfound.
- If it needs to be rewritten, the issues flagged by Wayfound are fed back to the
report-writersubagent so that it understands the mistakes made the first time. - This process continues until the report satisfies all guidelines.
Because Wayfound is a no-code agent supervisor platform, product and business owners (for example, the analysts at an investment firm) can specify the guidelines for the report generation using plain English. There is no need to modify the source code of the multi-agent system, which would require engineering resources.
Additionally, a complete record of every supervisor analysis performed is recorded in Wayfound. This provides a clear audit trail of an agent’s performance over time.
With the complete record of an agent’s performance, Wayfound can proactively suggest improvements that can be applied to the agent’s code base or knowledge system. These suggestions drive a virtuous cycle where the agent gets increasingly better at accomplishing its task. It is even possible to pull these improvement suggestions directly into agent coding tools like Claude Code and Cursor using the Wayfound MCP server.
At Wayfound, we are big fans of MCP and the new agent builders and SDKs. It feels like it is finally possible to create true multi-agent systems that guide themselves instead of following a pre-defined workflow. There will likely be a proliferation of agents in the coming months, and even with a solution like Wayfound to focus on each agent’s performance, like any idea or product, not all will be successful. But by leveraging a Wayfound supervisor agent via MCP, it is possible to dramatically improve the quality of the agent’s work and avoid the pitfalls of agent slop.
