Flo Crivello, founder and CEO of Lindy, outlines the company’s launch of “Lindy Teammate”, a Slack-based “AI employee” intended to work in a multiplayer setting across a team. He describes Lindy’s onboarding as connecting workplace tools and crawling full Slack history to build a knowledge graph, with separate personal and workspace layers and a background “memory agent” that refreshes context frequently. He argues that context, not raw intelligence, is increasingly the bottleneck for useful workplace agents, and says meetings should be treated as a first-class input into organisational memory.
A large portion of the discussion centres on agent operations and cost. Crivello says Lindy is again in “negative gross margin territory” after Teammate expanded what users ask the system to do, and he highlights caching as critical, claiming an 85% cache hit rate and warning that small changes can sharply increase costs. He describes internal engineering workflow changes driven by agents, including automated pull-request review and AI-led optimisation tasks, and predicts Lindy’s internal inference spending will overtake payroll within three to six months.
In the most politically contentious segment, Crivello argues for a sweeping US ban on Chinese frontier AI models, citing censorship and propaganda concerns, alleged distillation of US models, and national-security considerations. The host pushes back, suggesting alternatives such as insurance and auditing regimes rather than outright bans.
Commercial content includes an Anthropic sponsor read promoting Claude for personal “deep context” databases.
Asked about Lindy’s own internal use of AI, Crivello says inference spending inside the company is nearing payroll, and predicts it will exceed payroll within months. He also describes rapid changes to engineering workflow, including agents reviewing pull requests and staff increasingly “reviewing the machine that reviews the PRs”.
Read this moment →Flo Crivello, founder and CEO of Lindy, says the company is launching Lindy Teammate, an agent designed to sit inside Slack and act as a shared colleague rather than a single-user assistant. He argues this moves AI from “single-player” tools to “multiplayer” agents with shared team context.
Read this moment →Crivello says that when a company onboards to Lindy Teammate it can connect tools like Confluence, Notion, or Google Docs, but that Slack is “the most important.” He says Lindy then crawls the company’s Slack and builds a knowledge graph “based on the file system,” and that the graph has “2 layers”: a personal layer and a workspace layer.
Read this moment →Crivello says caching and context management are central to improving Lindy’s unit economics. He says Lindy’s cache rate is 85% and that small system changes can break caching, pushing it to 65% and nearly doubling costs.
Read this moment →Crivello argues that Chinese frontier models should be banned in the US even though Lindy uses DeepSeek by default. He cites alleged unfair distillation of US models, censorship concerns (including responses on Tiananmen Square), and the risk of Chinese systems acting as an “instrument of foreign propaganda” or operating “chunks of the American economy”.
Read this moment →