The Wartime CEO

The hosts open by celebrating Danny Postma discovering their podcast and calling Justin Jackson, Brian Castle, and Jordan Goff “legends” who had supposedly quit podcasting. Justin says the moment illustrates that founders can feel they are doing all the marketing while many people are only hearing about them for the first time; the practical lesson is to keep going. Brian jokes that the show’s biggest marketing problem is Justin’s absence from Twitter. Justin says he expects to return on November 28, prompting jokes about Elon Musk being indicted and Justin having insider information from a supposed relationship with Musk.

Jordan introduces Brian Halligan’s return as HubSpot CEO. Halligan and co-founder Dharmesh Shah announced that they were “completely refounding the company” and treating it as “day one,” despite HubSpot being a roughly 15-year-old public company with about $5 billion in ARR and a $10 billion market cap. Jordan’s first reaction was negative: bringing back a founder as CEO suggests something is wrong, not merely that the company wants a new leader for its next growth phase. The hosts nevertheless root for Halligan, whose personal wealth remains tied to HubSpot and who appears willing to return to the fight.

The hosts compare Halligan’s return with Ben Orenstein returning as CEO of Tuple. They distinguish the motivations: Orenstein seems to have come back partly because the company had become less exciting, while Halligan appears to be returning because HubSpot needs fixing. Orenstein acknowledged that Tuple’s performance was “getting a little bit soft,” which Justin describes as a threat to the company’s annuity. Brian says both founders may simply find building more fun than staying on the sidelines, although returning to run a public company sounds less attractive. Justin connects the pattern to Bob Iger’s return to Disney and the CEO’s desire to “roll up my sleeves” and be the hero.

The group argues that a founder’s greatest value may be product vision and taste: founders can decide to start from scratch or take an established product in a dramatic new direction, even if they are not the builders or designers. Halligan’s second announcement tweet focused on positioning HubSpot to “skate where the puck is going,” with the new tagline “HubSpot, the customer relationship harness.” The hosts credit HubSpot with defining and owning “inbound,” but Brian worries about the word “harness.” They interpret the announcement as an admission that simply being the dependable system holding all customer information will not be enough.

Brian says Builder Methods members increasingly build custom CRMs or case-management systems with AI, including a 50-person law firm and a dental practice. These nontechnical owners do not necessarily care about saving money; they want software that fits their industry’s unusual workflow instead of accepting the 40% fit of an off-the-shelf CRM. Justin cites Less Annoying CRM founder Tyler King, who said one customer left for Claude and later returned after discovering that the data was a mess. Tyler also argued that Salesforce may become merely a database connected through MCP, raising the question of whether vendors can charge $300 per user per month for database hosting.

The hosts debate whether internal tools have always threatened SaaS or whether AI makes the threat materially different by making custom software cheaper and faster. Brian says even 10% of an industry building its own tools could be massively disruptive, though some businesses will still lack product, UX, or technical skill. They see CRM as especially vulnerable because every small business believes its customer process is unique, unlike Slack, whose network effects and existing communities protect it. Justin ultimately changes his mind about Halligan’s strategy: HubSpot and other SaaS companies may survive by becoming customizable “harnesses” that let customers adapt the product rather than abandon it.

Jordan explains that Rosie users are asking for MCP controls so they can change settings from Claude. Rosie’s first response was to add a conversational box inside its own product—for example, letting a user type “I’m closing on Friday” instead of navigating several scheduling screens—but Brian argues Rosie should also provide an external MCP for Claude, Grok, or other agents. Being present in customers’ existing AI workflows offers convenience and marketing value, since an agent could remind them to configure something in Rosie. Anthropic’s announcement that partner apps can be embedded directly inside Claude makes the opportunity—and the danger—more concrete.

The hosts question predictions that NVIDIA-related competition will end the era of cheap AI tokens. Brian points to Anthropic’s S-1 and says its margins are already strong, not dependent on token costs being subsidized; he therefore expects intense competition and better models to keep prices from rising sharply. They are continually surprised that their current AI work costs about $200 per person. The major labs may charge businesses more, but Justin argues that attracting mainstream users requires free and low-cost plans. They admit the NVIDIA situation is unclear and increasingly political, but expect the labs to keep pushing prices toward the bottom for ordinary users.

The hosts criticize simplistic portrayals of activists opposing data centers, while acknowledging that the underlying disputes are more complicated. A proposed $5 billion data center on an empty parking lot could bring roughly $50 million a year to a small town, but residents may distrust local governments after earlier deals failed to deliver promised schools, clean water, or jobs. Jordan notes that some large companies have sued communities after residents voted against projects. The group says AI companies should win local support through subsidized electricity, investment, and enforceable commitments rather than grand predictions that money will disappear or that everyone will simply pursue hobbies. Anthropic’s frightening claims about job loss and existential risk also make constructive discussion harder.

Justin describes an “Amp Week” experiment inspired by Quinn Slack and Thorsten Ball’s Raising an Agent podcast and the operating ideas of 37signals. Transistor moved its team meeting from Thursday to Monday; Justin and John now discuss priorities on Thursday, send a Sunday-night email identifying the week’s important work, and put those priorities into a shared Google Doc. During Monday’s meeting, each person spends five minutes choosing what to work on rather than reporting by department. Justin says this replaces old product, engineering, and customer-success headings with a single question: what matters now? Brian describes a similar Tuesday stand-up using Linear to move, defer, and clarify individual tickets.

Brian announces two related experiments. Builder Methods now offers ordinary coaching calls plus “project advisory,” a 30-day partnership with up to three live calls and unlimited asynchronous messages. The goal is to work alongside a customer building an app or agent system, rather than trying to accomplish meaningful work in a single 60-minute call; it also gives Brian direct exposure to how companies are using AI. He is separately developing Agent Reports, a platform where agents brought by the customer can perform scheduled jobs and report completed work, blockers, and approval requests to a team. His larger thesis is that companies will have both people and agents with defined jobs, requiring reporting infrastructure instead of a chat window for every task.