We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Big Ideas
Agents are crossing from assistance into commerce—and their distribution may be social. Instinct says purchasing users spend more than $1,300/month on average through the product and is partnering with Stripe; its examples include international travel, groceries, appointments, and price discovery. Grok Bot now completes online purchases after a user connects Link. At the same time, bots are becoming shareable artifacts: one student shared copyable, tailor-able bots for chief-of-staff, job-finding, and PM workflows, while Hiten Shah argues that “Do you have a bot for that?” could make people the discovery layer.
For PMs, this shifts the core loop from ask and answer to authorize → execute → recover → share. Instrument completed jobs, repeat usage, payment failures, and referral activation—not just chat volume.
Context is a delegation metric. A Company OS approach pairs OpenClaw’s Slack/Telegram gateways, scheduler, and persistent identity with Hermes, which turns repeated requests into skills. Its “product context coverage” test scores industry, business, and customers; it forces five unknowns per area, counts only knowledge retrievable without guessing, and prioritizes the three gaps that raise coverage fastest. The guide’s example treats 54% as suitable for backlog decisions and 70–90% as a target for strategy-level work. Use coverage as a gate: delegate drafting at low coverage, but require human review for decisions until context, permissions, and evidence are sufficient.
Tactical Playbook
Treat early churn as a research signal. RemindMe’s builder had four real users stop, no response, and no business-owner interviews. A practical sequence from the thread: (1) ask former users what they were trying to do when they opened it—or what they use instead now; (2) speak to problem-holders without pitching; (3) run the B2B confirmation log manually for two businesses for a month. If nobody wants the manual service, stop adding software.
Make design critique about fit, not taste. Separate craft—spacing, color, wording—from fit—whether the screen survives the real workflow. Tie choices to discovery-defined jobs and prompt/ability/desire; test edge states such as 4,000 rows, read-only roles, failed syncs, and empty accounts.
Case Studies & Lessons
A purpose-built renewal agent made proprietary context scalable. The team avoided generic sales agents for renewals because they lacked account-specific data; its sub-agent combined Salesforce contracts, LTV, and engagement with social, podcast, website, and Gmail context, then generated branded decks through Gamma. It extended comparable collateral to every sponsor, including smaller accounts, and produced 20–30 customized pitches versus roughly two before AI.
The operating pattern matters more than the deck: use a generic agent for re-engagement, collect feedback, have the specialized agent propose a narrative, review it with a human, then generate the collateral. The episode reports an A/B test in which this sequence worked better than one-shot customization. The agent still invented numbers despite explicit instructions, so numerical guardrails and human review remain part of the product.
Agentic coding needs gates, not just speed. Slack Code creates a project channel with the agent’s plan, line-by-line diffs, and live preview; participants can pause, redirect, or stop it, and production still requires human signoff. The episode cites a 2026 report finding exploitable vulnerabilities in roughly 44% of AI code-generation tasks even as AI-assisted developers committed code 3–4x faster. For PMs, use the workflow as a prototype-contribution model—tag an engineer before approval—and keep standard pull-request/release gates, permission checks, and reviewer ownership.
Career Corner
Build synthetic experience, not just a title. Aakash Gupta’s ladder is courses → a model-using product at a live URL with evals → real users → freelance/consulting → shipping AI as a PM. He says only one in eight jobs require production AI experience, so Levels 1–4 can create evidence before a formal AI-PM role. The practical artifact is a live workflow with users, evaluation results, and a decision log—not a certificate alone.
Tools & Resources
Product Quest is a community-recommended practice platform built around realistic scenarios in prioritization, discovery, metrics, strategy, and stakeholders, with learning paths and simulated career progression. It is worth testing as deliberate practice for interviews or a transition—not as a substitute for shipping.
