I Email All My Bots

Jordan and Brian begin with Rob Walling’s recently published MicroConf talk about the changing risks facing SaaS companies. Rob reportedly listed roughly ten different dangers, including a product becoming obsolete because customers no longer need the underlying work, or a platform a company depends on launching a competing feature. His stated goal was to scare founders “but not like too much”: the talk was intended as an alarm bell, followed by a more positive discussion of how to respond. Jordan says the practical conclusion is that founders should operate with some paranoia. Even if a business currently feels safe because of its niche, technology and markets can change in unexpectedly large leaps. They use a hypothetical receptionist product as an example: if an AI system such as “Rosie” eliminates much of the receptionist’s work, demand for receptionist software could disappear. But Rosie itself is also vulnerable because competitors can build similar products quickly. Brian asks whether Rob’s warnings apply equally to startups and established SaaS companies. They agree that both face risk, but the risk is different: a new founder is deciding what to build, while an established product has customers, revenue and a more defensible position.

The conversation then turns to Einar Vollset’s adjacent MicroConf talk about moats and what makes a software company valuable. Jordan describes Einar’s material on the Discretion Capital website as a current guide to modern moats, informed by Einar’s work evaluating and selling software businesses to buyers such as private-equity firms. Together, Rob’s talk is framed as the company-building side of the issue, while Einar’s is the liquidity and sale-value side: what remains valuable when products can be built and copied much faster. Brian says he has mostly stopped trying to prescribe one universal way to build a SaaS company because the answer depends heavily on the category. He is much more focused on deciding which products are worth starting. The old bootstrap pattern—turning a painful spreadsheet or repetitive process into a narrowly targeted SaaS—now looks less attractive because a founder can often build a private version of that tool in minutes with an AI coding assistant. Jordan gives his old idea of a metrics dashboard as an example: he would not build it as a standalone product today because he could make his own version quickly. They also note a second risk: OpenAI or Anthropic may already be building, or may soon build, the same product. Vertical SaaS for less technically sophisticated industries may still offer opportunity, but they suspect founders may have only a few years before that advantage erodes too.

The speakers argue that the uncertainty strengthens the case for bootstrapping or using only non-institutional funding at the beginning. A new product may turn out to be a major company, a small business, a side project, or a short-lived income source, and it is difficult to know which outcome is coming. Raising substantial capital before that is clear can leave a founder committed to a business whose useful life is shorter than expected. They revisit Josh Pigford’s AI-powered SEO service as an example. Josh has reportedly reached roughly $470,000–$500,000 in ARR while operating the service largely by himself with AI, rather than recreating the 30-person team behind Jordan’s former Audience Ops. Brian thinks AI-enabled, productized services still have room to grow, including services that help companies adopt AI. The key framing from Einar’s talk is to sell the outcome rather than the tool: customers care about the SEO result, not logging into another application and clicking around. Jordan remains cautious because marketing services are paid from marketing budgets, which are easy to cut. A service can perform well and still be brought in-house. He also questions whether the many new “AI automation” consultants and agencies are genuinely durable, since services retain messy client-specific needs. Still, both hope Josh’s model proves that a very small team can use bots to deliver a valuable outcome.

Brian asks what Jordan is doing with automation. Jordan describes his current operating model as being effectively a company of one, with one person helping run Clarity Flow. He wants to do two things continuously: build one product—currently a video-creation product—and keep marketing systems running for the rest of his portfolio. His earlier weakness was launching products, getting an initial audience-driven spike, and then allowing them to sit idle. The new system is intended to give each product a continuing chance to grow. Jordan uses Grok Bot as a team and job runner rather than as a chat-based personal assistant. He has a chief-of-staff bot and roughly five or six individual bots, with each bot responsible for one product. They perform recurring work such as SEO foundations, metadata and sitemaps, AI-visibility work, competitor-comparison pages, guides and free tools. He also connects each bot to Agent Mail, giving every bot an email address. Rather than opening a chat interface, Jordan receives reports and approval requests by email and replies with concise instructions such as approving items one and two, rejecting item three, or asking for a revision. The bots also propose marketing experiments. For Flywrite, for example, a bot can research competitors, suggest experiments such as comparison articles or cold outreach, create the relevant tasks after Jordan approves one, and later report what the experiment taught them. Jordan says he spends his time designing and refining the skills, jobs and approval emails so that his involvement takes about five minutes while the systems continue working.

Brian focuses on the unusual division of labor in Jordan’s setup: Grok Bot manages a team-like structure and recurring jobs, while Claude performs the substantive work. Jordan says Grok Bot is essentially a scheduler and job runner for him, not the model he relies on for writing or analysis. Whenever a bot needs to do meaningful work, it starts sessions in his Claude account. Jordan has a Claude Max plan and wants important work handled by Claude Opus 5.5, which he says he trusts and considers the best model he has used. They compare this approach with the growing number of agent products from major AI companies. Jordan mentions Claude’s newly launched agent or team functionality, Meta’s Muse, OpenAI’s dots, and other tools, while arguing that open-source or hackier systems such as OpenClaw and Hermes are too complicated and unreliable for his needs. His preference for Grok Bot is specifically its team metaphor: he can have multiple specialized bots and a chief of staff rather than one general assistant. Brian observes that Jordan has confidence in Grok Bot for managing tasks but confidence in Claude for the actual work and writing. Jordan says he may revisit Claude’s own offering, but he currently values the multiple-bot setup. He also emphasizes that the system is still being improved: early reports contained too much information, so he has repeatedly redesigned the emails to make approvals fast on an iPhone. The goal is not full autonomy; it is efficient human oversight.

Jordan says the marketing automation is valuable because it keeps products moving after he has shifted his attention to something new. SEO may produce only small amounts of traffic for three, four or five months before Google or ChatGPT suddenly begins sending meaningful traffic. His bots provide the ongoing investment needed to reach that point. He cites Cory Haynes as another example: Cory reportedly has 20 or 30 apps, and even slow growth is useful if the line is at least gradually moving upward. A product does not have to be the founder’s only focus to remain a potentially growing asset. This system also supports Jordan’s preferred strategy of not going all-in on one bet. His one major commitment is to staying small, continuing to build, and making the overall portfolio profitable and sustainable. He says he spent several weeks in an unhealthy cycle of asking what product he should build next and worrying that he might make the wrong bet. He finally broke out of it by starting the video-creation product. The work itself reminded him that he needs to be building, whether that means a product or an automation system. Jordan’s conclusion is blunt: he “creates his way out” of these funks. Every product idea still carries risks—AI could make it irrelevant, or the marketing system might not work—but he can build, launch and learn within a month or two. Brian agrees that action is what unblocks excessive analysis. Jordan also says his personal goal has changed: he would welcome a huge exit, but he is now more motivated by a comfortable lifestyle, useful work and the craft of building.

Jordan introduces his friend Jared, who has operated Appointlet, a scheduling product, for roughly 12–14 years. Brian remembers using Appointlet to schedule Restaurant Engine sales calls. Jared recently launched Little Desk, an adjacent scheduling product specifically for parent-teacher conferences. Jordan says the important asset was not necessarily Appointlet’s old codebase—Jared told him the code was the least relevant part—but the accumulated knowledge of running a software company, including its lessons about customers, onboarding, marketing and operations. Jordan sees Little Desk as an example of how faster building changes the cost of pursuing a focused niche. Rather than adding another industry and feature set to a horizontal product, Jared could create a separate, simpler product around a customer segment he understood. Little Desk has fewer features and much simpler pricing. It charges annually, which Jordan thinks may make the economics of advertising work better: the initial payment can cover a significant portion of acquisition cost, whereas a low monthly plan may not provide enough runway. Brian strongly likes annual pricing. Apart from Clarity Flow, his products are generally annual: Builder Methods Pro, Flywrite at $99 per year, Mimio at $500 for the first year, and the video tool he is building as an annual Mac-app license. Annual billing gives him revenue upfront and creates a meaningful commitment for the customer. They note that monthly SaaS customers may leave a $19, $29 or $49 subscription running for several months without logging in, then cancel. Jordan has considered offering Builder Methods Pro customers a monthly downgrade only when they try to churn, as a possible retention tactic.

The annual-pricing discussion expands to Jordan’s infrastructure costs. He is considering moving LiveKit to quarterly billing, but describes its existing annual-commitment model as unusually fair. LiveKit estimates usage, provides credits at a discounted rate through its infrastructure and model agreements, and divides the annual commitment into equal monthly payments. If the customer uses the credits faster than expected because the business grows, the monthly payment does not suddenly increase; the agreement simply ends or renews sooner. Jordan says their previous agreement consumed its credits in six months, effectively at twice the expected rate, while their pricing remained unchanged during those six months. Brian then enthusiastically recommends Claude Opus 5.5. He says an earlier Opus release was expensive and unpopular, leading him to use other models and Codex. In July and August, his AI usage reached roughly $1,000 per month. After returning to Claude’s $200 Max plan with Opus 5.5, he says he has been multitasking heavily, including building products and operating bots, without coming close to the usage limit. He calls it both the best model he has used and unusually affordable for the amount of work it handles, adding that he almost hopes Anthropic does not change it. They finish with AI-company speculation. Brian discusses Walter Isaacson’s recent interview with Elon Musk, in which Musk reportedly said he regretted spending a year on political work rather than focusing on xAI earlier. Brian thinks xAI should not be overlooked because it has SpaceX’s resources and compute, even though it started later. They also react uneasily to Musk’s claims about money, simulations and robots controlling the future, and mention Palmer Luckey and a reported startup developing genetically engineered military dogs. The episode closes with weekend plans, including date night, children’s activities and soccer.