Solo Founder Manages 15 AI Agents, Cutting Slack

Ryan Carson's monthly expenditure on AI agents, specifically Devin, saw a substantial increase, at one point reaching $20,000. This growth in operational costs is directly tied to the expansion of his business, Untangle. Initially, his investment in Devin agents was around $5,000 per month, but as the platform's capabilities were integrated more deeply into the business operations, the costs escalated. Carson emphasized that effectively managing these agents has become a critical skill for modern founders.

Carson has fundamentally changed his approach to managing AI agents, moving away from a hands-on, prescriptive method. Previously, he would interact with an Integrated Development Environment (IDE) to provide detailed instructions to the agents. Now, his strategy involves assigning broad goals and allowing the agents to operate independently, without continuous human oversight. This shift enables him to manage 15 concurrent agents effectively.

Untangle's initial concept involved using AI to assist individuals with divorce proceedings. However, direct market validation revealed that consumers preferred human lawyers for such sensitive matters. This feedback prompted a significant pivot towards a B2B SaaS model, targeting family law firms. The company now focuses on addressing paralegal shortages within these firms, providing a more relevant and in-demand solution.

Ryan Carson initially experimented with cloud agents and was surprised by their maturity and readiness for production work. This realization prompted him to commit heavily to these tools, leading to a monthly expenditure of $5,000 to $20,000 on Devin agents. His successful implementation and demonstration of platform usage eventually led to obtaining credit support from Cognition, underscoring the potential he saw in cloud-based AI for future operations. Carson firmly believes that the future of AI is predominantly with cloud agents, anticipating a nearly 100% adoption rate.

Carson strongly recommends using AI agents to streamline the process of drafting investor updates. These agents can also be instrumental in tracking and compiling critical business health metrics, ensuring that founders can provide timely and accurate information to their investors. This approach frees up valuable founder time, allowing them to focus on strategic growth rather than manual data aggregation and report generation.

Carson highlights that experienced founders possess a significant advantage in orchestrating AI agents due to their developed management skills. He asserts that expertise in team structure and delegation, honed over years, directly translates to effectively managing AI agents. This perspective suggests that seasoned professionals are particularly well-positioned to leverage AI, as their established leadership capabilities can be directly applied to these new technological tools.

Ryan Carson employs a time bucketing strategy, prioritizing the allocation of his time over managing individual tasks. He uses decision fatigue as a heuristic to identify and deprioritize features that lack strong conviction. If a feature feels like a burden to decide on, it often signals that it should be allowed to 'die on the vine,' indicating a lack of true importance or commitment.

Carson expresses a critical view on the growing reliance solely on digital product building, arguing that many founders are not engaging enough with real people. He stresses the paramount importance of direct customer engagement to discover true product-market fit. He advocates for founders to physically visit offices and speak with users in person, emphasizing that genuine insights often come from these real-world interactions rather than purely online feedback.

Carson challenges the common perception that cloud coding agents are exclusively for writing code. He highlights that AI platforms are evolving into broader business agents capable of handling diverse operational tasks. For instance, he utilizes Devin not just for coding but also for critical business functions such as Deal Desk management, generating quotes, creating documentation, and customer triage, demonstrating a versatile application of AI beyond its traditional development roles.

Ryan Carson differentiates between local and background agents based on their optimal use cases. He uses local agents, such as Codeex, for high-latency tasks that require interactive pair programming or managing large feature initiatives. Conversely, background agents are deployed for automated tasks like bug autofixes, one-off assignments, and general operational support. He also noted a preference for specific agents dedicated to front-end UI grinding and verification tasks, indicating a nuanced approach to agent selection based on task demands.

Ryan Carson has implemented a 'land PR' playbook within Devin to ensure high code quality before merging pull requests (PRs). This automated process triggers a fresh Devin review on the PR and generates a video walkthrough of the proposed changes. The video output allows for quick human approval, significantly streamlining the verification process and maintaining code integrity with minimal manual intervention.

Ryan Carson is actively working to eliminate Slack from Untangle, arguing that it fragments context and hinders efficient communication. His plan is to mandate that all multi-user interactions occur directly within Devin, centralizing communication around AI agent activities. However, the interlocutor presented a contrasting viewpoint, suggesting that public Slack threads are effective for transparently showcasing AI capabilities to team members and fostering broader adoption.

During his tenure at Launch Darkly, Carson found that deploying background agents in public Slack threads served as a powerful 'change agent move.' This strategy fostered transparency, helping employees understand and integrate AI into their workflows, ultimately improving human-AI collaboration. The broader goal is to transform Slack culture so that every thread directly translates into tangible work or action, reinforcing the utility and direct impact of AI agents within the organization.

Ryan Carson employs a dual-tool workflow for building technical design systems. His process begins with using Claude to create initial design files. These designs are then downloaded and imported into Cursor, where they are combined with existing Figma assets to construct the technical components of the design system. This integrated approach effectively bridges the gap between design conception and development execution, streamlining the creation of robust design infrastructures.

Carson utilizes OpenDevin for managing background agent tasks, indicating its role in automating continuous operations. For active coding and email management, he expresses a strong preference for Cursor, particularly praising its user interface. He highlighted Cursor's efficiency by noting its recent success in processing 2,000 unread emails, showcasing its utility for both development and organizational tasks.

The hosts underscored that merely generating extensive AI code does not guarantee the creation of a superior product. They emphasized that direct, real-world human interaction with customers remains fundamentally critical for product success. They advise founders to strategically employ AI tools for background tasks, while maintaining a primary focus on understanding and engaging with their users.