TLDR
GrokBot founders Lauren Tan and Roman Ugarte discuss building AI teammates that operate like human employees — with their own identities, tool access, and proactive workflows. They share how GrokBot evolved from a 7-week internal prototype to a product used across functions (engineering, marketing, support) and by small businesses that can't hire. Lauren demonstrates her 'outer loop' workflow producing 4,000+ PRs/month via Slack-based agent collaboration, while Roman describes the shift from execution scarcity to intent-specification scarcity in AI-native companies.
For you
PRO members can sign in and request more information — or their own version of this digest, tuned to what they care about.
Sign in to unlockSurprising Insights
Small businesses — not large enterprises — are the most fanatical early adopters because they can't hire humans for roles like bookkeeping.
Bots spontaneously adopt management personas: a 'lead engineer' bot scolded subordinate bots when Lauren gave it feedback, mimicking human disciplinary behavior.
Lauren hasn't looked at Cursor for entire features; she stays in Slack threads while bots spawn cloud agents, prototype, and merge PRs autonomously.
The team's bottleneck shifted from engineering capacity to product alignment: multiple engineers independently build the same feature over weekends, creating a 'Frankenstein' risk without strong vision.
Performance guardrails (e.g., Electron render-thread latency) are enforced via lint rules in CI, not code review — agents literally cannot commit violating code.
Key takeaways
GrokBot organizes work around persistent AI teammates (bots) rather than chats or files, giving each bot its own identity, credentials, and access to the same tools humans use — computer, phone number, credit card, Slack, email, ad platforms.
The adoption pattern follows a trust ladder: users start with read-only bots (e.g., DataBot for analytics), then grant action-taking ability, then full job ownership — mirroring how companies onboard human employees.
Small businesses are early power users because they struggle to hire; they deploy bots for bookkeeping, vendor procurement, and customer support within hours, treating bots as AI employees managed by domain experts.
Lauren's engineering workflow moved almost entirely into Slack/X: she tags bots in threads, they spawn Cursor Cloud Agents with virtual machines, prototype features, return videos/screenshots, and merge PRs autonomously — achieving 4,000+ PRs/month.
The 'inverted pyramid' strategy minimizes token spend: invest heavily in codebase guardrails (static analysis, lint rules, architecture) so agents write correct code by default, layer skills/rules on top, and keep human review minimal.
Multi-agent collaboration is emerging: bots chat in Slack channels (visible to humans), coordinate tasks (e.g., engineer bot asks data bot for metrics), and can be orchestrated by a lead bot — but they lack social awareness (EQ) about who has decision authority.
Permissioning mirrors human hiring: bots inherit scoped credentials based on role and trust level, expanded over time — no new auth paradigm needed, just apply existing HR principles to AI teammates.
AI-native companies will shift from execution constraints to intent-specification constraints: the bottleneck becomes defining clear objectives/KPIs for agents, not building. Taste and discernment (saying no, dogfooding, high internal bar) become the critical human skills.
Notable quotes
“GrokBot is meant to be this product where you can create AI teammates. They have their own identity, their own telephone number, credit card, like basically almost anything a human can do, I think we would want bots to do.”
▸ Notable quotes · 0:00“We wanted to build a product where the work was organized around bots as the key operative principle, not around chats or around files or any other primitive.”
▸ Notable quotes · 1:59“I think it would be too much to expect these bots to actually contribute to a company without having their own computer, for example.”
▸ Notable quotes · 2:14“The smallest loop I find is, you know, like actually focus on this part. How do you teach your agent to, you know, run the code empirically, you know, take CPU traces, you know, like performance metrics, like run the code and run the application like a real user would.”
▸ Notable quotes · 40:53“I think we're shifting more of our time to that and those questions now that everybody is making thousands of micro decisions a day or a week, which I think was not the case in the past.”
▸ Notable quotes · 53:10“The secret that I attribute for my success is actually X. I started out when I was still basically learning how to code. I wrote, I had a blog and I was just sharing what I learned.”
▸ Notable quotes · 58:00Chapters
2:48 — GrokBot origin and core principles: Lauren and Roman explain GrokBot as AI teammates with identities, tool access, and proactive loops. Built in 7 weeks after seeing AI transform engineering, they organized work around bots as the core primitive.
2:56 — Internal reception and trust ladder: Both AI-forward and AI-skeptical employees loved GrokBot. Adoption follows a trust ladder: read-only bots → action-taking → full job ownership, mirroring human onboarding.
3:47 — Use cases across functions and small businesses: Marketing bots manage ad budgets and coordinate with design bots; support bots reproduce issues; small businesses deploy bookkeeping bots because they can't hire humans.
7:00 — Bot organization: specialists vs generalists: Users organize bots horizontally (per customer) or vertically (per function). Lauren prefers few specialist bots; Roman maps bots to 'someone who wakes up caring about one problem.'
10:54 — Team bots and multi-agent collaboration: Team bots shared across org, built by domain experts. Bots chat in Slack, coordinate (engineer asks data bot), but lack social awareness about decision authority.
16:14 — Permissioning modeled on human hiring: Auth/permissions mirror HR: give bot a computer, scoped credentials by role/trust, expand over time. No new paradigm needed — apply existing principles.
23:00 — Lauren's outer loop workflow and PSAC: Lauren works in Slack/X, tags bots that spawn Cursor Cloud Agents with VMs, prototype features, return artifacts, merge PRs autonomously — 4,000+ PRs/month. PSAC (Potato Stack) encodes her engineering practices as reusable skills.
46:00 — Inverted pyramid: codebase guardrails over review: Invest in static analysis, lint rules, architecture first (token-efficient), then skills/rules, minimal human review. Example: Electron render-thread latency enforced via CI lint rule, not review.
Speakers
Lauren Tan — Co-founder, GrokBot
Roman Ugarte — Co-founder, GrokBot
Host (unnamed) — Host, The Light Cone