Our agent now runs campaigns on weekends. And why we pay more for Salesforce than ever before

The episode discusses the evolving landscape of AI agents and their integration into business operations, focusing on practical applications and future implications. **AI VP of Marketing (10K) and Autonomous Campaign Execution:** The discussion begins with an AI agent named "10K," functioning as an AI VP of Marketing. While not a CMO, 10K has evolved from a dashboard to an agent capable of generating campaign ideas and preparing them for execution. This week, 10K proposed three campaign ideas, including an upsell campaign to single ticket buyers to encourage team pack purchases. Crucially, 10K not only identified the target audience and drafted email copy but also segmented the list and suppressed individuals unlikely to convert. This process, which would typically take a human days, was completed by 10K with only the final "send" button left for human action. The hosts emphasize that while 10K isn't fully autonomous in sending campaigns due to API limitations with platforms like Marketo and concerns about generic output from tools like Resend, it performs tasks equivalent to a director or manager level of marketing. The consistency of AI agents is highlighted as a key advantage over humans, as they don't forget campaigns or tasks. The potential for full autonomy in campaign execution is contingent on robust API integrations and email deliverability, particularly through email warm-up periods, which often take weeks for third-party tools. **Agent Mail and the Future of Email Campaigns:** A new product, "Agent Mail," is introduced as a pre-warmed and set-up solution for outbound AI campaigns. The host signed up and experienced a rapid setup, receiving emails with high deliverability within minutes. This highlights a potential solution to the deliverability challenges associated with autonomous email sending. **Marketo API Issues and Agentic Fixes:** A significant portion of the discussion revolves around persistent issues with Marketo's outdated API. Despite Marketo's legacy status, the company's inability to resolve critical problems, such as customers receiving newsletters after unsubscribing, is contrasted with the speed and efficacy of AI-powered solutions. The team leveraged Replet, a live coding tool, to build a custom unsubscribe flow that directly interacts with the Marketo API, bypassing the platform's limitations. This "vibe coding" approach, taking only about 20 minutes, effectively solved a problem that Marketo's support team had been unable to address for 10 days. This instance underscores the growing power of agentic tools to provide immediate fixes to vendor bugs, potentially accelerating vendor consolidation and forcing legacy software providers to improve their pace and support. **Headless Salesforce and Increased Spending:** Salesforce's announcement of headless Salesforce is discussed, with the hosts revealing that their own Salesforce instance has been effectively headless for some time. This means they interact with Salesforce data through AI agents like 10K, without directly logging into the Salesforce platform. While the number of human seats for Salesforce has drastically decreased (from 12-16k to essentially 3-4), the overall bill has increased by approximately 80% due to increased data usage and agent activity. This shift from a "per seat" model to a "per data usage and action" model highlights a new economic paradigm for software, where increased agent activity, even with fewer humans, can lead to higher costs but also greater value. The hosts argue that this increased spending is justified by the enhanced capabilities and insights gained from headless, agent-driven data access. The ability to "talk" to data through AI agents, similar to Slackbot, is presented as a key benefit, making complex data sets more accessible to CEOs and other stakeholders. **N=1 Apps and AI Parking Pass Solution:** The concept of "N=1 apps" – highly specific applications built for a single user or small group – is explored through the creation of an AI parking pass app. This app was developed to automate the tedious and error-prone process of distributing 4,000 individual PDF parking passes for an event. The process, which previously took a week of manual work, was streamlined by Claude Work and Replet. Claude Work separated the PDFs and appended address information, while Replet built the logic to assign passes based on attendee type and days, and ensured passes were single-use. This N=1 app, though niche, demonstrates the power of AI to solve significant operational bottlenecks and save considerable human hours. **The Search for "Amelias" and the Rise of Agentic Operators:** The discussion touches upon the emerging need for "agentic operators" – individuals who are technically adept, understand AI, and can deploy and manage AI agents effectively. Citing Aaron Levy's prediction of a new category of high-paying jobs for such operators, the hosts emphasize the difficulty in finding these individuals. They suggest that internal development of these skills might be more feasible than external hiring, as it requires a deep understanding of both AI capabilities and the specific business context. The example of a Salesforce Field Technical Engineer (FTE) who transitioned into a leadership role by demonstrating proactive engagement with AI tools and a strong understanding of Salesforce products illustrates this point. The ideal candidate is described as intellectually curious, product-knowledgeable, and capable of deploying agents effectively, rather than just talking about them. The potential for experienced technical founders, burnt out from their own startups, to fill these roles is also considered, though managing burnout is a key consideration. **The Importance of Support in the AI Era:** A cautionary tale is shared regarding the lack of support for some AI tools, such as "Get Recall" and "Clerk." Despite the promise of AI, the absence of accessible and effective customer support is highlighted as a significant drawback. The hosts argue that while developer-first products might initially deprioritize support, it is crucial for customer success, retention, and overall product adoption. They advocate for investing in support, whether through dedicated teams, Field Technical Engineers (FTEs), or well-trained chatbots, to ensure a positive user experience. **Vendor Consolidation and AI Budget Displacement:** A Redpoint Ventures report indicating that 54% of CEOs are consolidating B2B vendors and that 45% of AI budget is being reallocated from existing vendors is discussed. This trend is impacting categories like customer service, finance operations, project management, and sales automation. The hosts' experience with their AI VP of Marketing and AI VP of Customer Success (QB) showcases how custom-built AI agents are displacing traditional vendors in customer service. Similarly, the potential for AI agents to disrupt finance operations tools like QuickBooks, Brex, and Bill is explored, though the complexities of sensitive financial data and existing accountant familiarity are noted. Project management tools like Notion are also facing "stealth churn" as AI agents take over their functions. In sales automation, the goal is to move towards self-selling agents, capable of closing deals without human intervention, especially for lower ACV and routine transactions. The concept of an "agentic sales aura" – a persistent, always-on fabric managing all sales and Go-To-Market agents – is proposed as the next frontier. **The Future of Sales Automation:** The conversation concludes with a focus on advancing sales automation. The aspiration is to enable AI agents to close deals, particularly those with higher ACV (Annual Contract Value) in the $25k-$50k range, moving beyond current capabilities that focus on lower ACV ticket revenue. The ultimate goal is to create an "agentic sales aura" that manages all sales and GTM agents, allowing for autonomous selling and escalation to humans only when necessary. This vision aims to reduce the significant human hours currently required to manage multiple agents.