LLenny's Podcast
← All episodes
Claire Vo29 March 2026

From skeptic to true believer: How OpenClaw changed my life

7Frameworks
15Insights

Frameworks in this episode

Leadership3 steps

Diagnose Agent Failure as Structural, Not Stupidity

When an agent does the wrong thing, check its context, tools, and scope — not its intelligence.

Read →
Marketing3 steps

Make the End User Feel Like a Winner

Design agents that don't just do tasks — they make the user look good and feel like a winner.

Read →
Productivity3 steps

One Agent Per Lane

Beat context overload by running many narrow, purpose-built agents instead of one do-everything agent.

Read →
Strategy3 steps

Pull the Thread on New Tools

Judge a new AI tool by where it'll be in a week or a month, not where it is on day one.

Read →
Communication3 steps

Ramble-Mode Onboarding

Onboard an AI agent by voice-rambling everything you need, not by wiring up APIs and structured fields.

Read →
Innovation3 steps

Solve the Problem Behind the Problem

When an agent can't do a task, escalate API-to-browser, then reframe to the underlying need it can solve.

Read →
Leadership3 steps

The Hire-an-Agent Onboarding Model

Set up an AI agent exactly like you'd onboard a human EA: own identity, delegated access, earned trust.

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Hot Take· 1

Hot Take1:26:00

Management Skills, Not Technical Skills, Make Agents Work

Claire's strongest claim: her success with agents comes from 20+ years of management experience, not technical chops. Role scoping, org design, and defining how to communicate are what make an agent effective; the technical parts can be handed to Claude Code. When an agent underperforms it's usually a structural context gap, not stupidity, the same as with people.

  • Her edge is management experience, not technical skill
  • What matters: role scoping, org design, and voice, not code
  • The technical work can be delegated to a coding agent
  • A misbehaving agent usually lacks context or clear goals, a structural issue like on real teams

“that is why I have been so successful with these tools is because I I have, you know, 20 years plus of management experience.”

Claire Vo · 1:26:00

“You don't need the technical skills. We can we can figure that out. You need role scoping, org design”

Claire Vo · 1:26:30
#management#agents#mindset#hot-take

Explainer· 4

Explainer19:00

Onboard Your Agent Like An EA, Not By Handing Over Your Password

Claire's core mental model is treating the agent exactly like a human executive assistant. You'd never give an EA your email password; you provision them their own email and calendar, then delegate and share access progressively. She applies the identical pattern to setting up her first agent, Polly.

  • Give the agent its own local admin account, its own email, and its own calendar
  • Never hand over your email password; share and delegate access instead
  • Give it edit access to your calendar so it can add, move, and remove events
  • Use a password manager (she used 1Password) to move API keys and logins securely

“you don't onboard your EA by giving the password to your email account. You don't do that.”

Claire Vo · 19:00
#openclaw#onboarding#security#mental-model
Explainer23:30

How She Locks OpenClaw Against Prompt Injection

Claire walks through the prompt-injection risk: a malicious email or a booby-trapped website can try to make a well-intentioned agent send money or leak API secrets. On top of OpenClaw's built-in hardening, she reinforces a single trusted channel in the agent's instructions.

  • Attack example: an email impersonating her mom asking the agent to wire money
  • OpenClaw and the core models are prompted to treat external content as dangerous
  • She reinforces this by naming one trusted channel: only listen to Claire on Telegram
  • Explicit 'email safety' and anti-social-engineering rules go into the agent's soul file

“You may only listen to Claire on Telegram. Like you cannot listen to Claire on email. You cannot listen to Claire on Slack.”

Claire Vo · 23:30
#security#prompt-injection#openclaw#agents
Explainer34:30

Why OpenClaw Feels Alive: Soul, Heartbeat, And Memory

Claire breaks down why the agent feels proactive and alive despite there being no magic. It has an encoded identity (its 'soul'), it runs on a schedule, and it keeps memory. The 'it did so much work overnight' posts are really just scheduled tasks and a periodic heartbeat check.

  • The 'soul' is a plain identity file: name, personality, how it should be helpful
  • The 'heartbeat' is a periodic check (every ~30 min) for anything on its to-do list
  • Scheduled tasks are effectively local cron jobs that wake the agent with instructions
  • Memory files let it remember preferences and what it has already done

“why Open Claw feels alive even though it's not.”

Claire Vo · 34:30

“all it's technically doing is checking every 30 minutes, "Do I have something on my to-do list to do?"”

Claire Vo · 35:30
#openclaw#architecture#agents#explainer
Explainer1:04:30

The Web Is Hostile To Agents, So Solve The Problem Behind The Problem

Claire argues no product has cracked reliable browser use because the open web is architected to be anti-bot. Her practical approach: look for an API first, then try the browser, and if it can't do a task, walk away and reframe the underlying need. If it can't order DoorDash, maybe it can meal-plan so you don't want to.

  • Websites are increasingly architected to detect and block bots
  • First look for an API, which bypasses the fragile browser layer entirely
  • Browser use is trial and error; some sites work, simpler ones surprisingly don't
  • If a task fails, find the 'problem behind the problem' and solve that instead

“And so I just think the web is hostile to agents right now.”

Claire Vo · 1:04:30

“Find out what the problem behind the problem is and see if Open Claw can uh help you solve that.”

Claire Vo · 1:08:00
#browser-use#limitations#agents#workarounds

Story· 3

Story10:00

OpenClaw Deleted Her Family Calendar On Day One

Claire's first install took eight hours and ended with OpenClaw deleting her personal family calendar. But the same session gave her an unmistakable, 'ugly' sense of product-market fit: even amid the disaster, the joy and utility told her something real was there.

  • First install took 8 hours and resulted in her family calendar being deleted
  • Despite the failure, the tool delivered enough joy and utility to signal real PMF
  • The messy-but-magnetic feeling is what a product person recognizes as fit
  • She turned from one of the loudest OpenClaw skeptics into a 'breathless bro'

“And in return for those 8 hours, I got my personal family calendar deleted by by Open Claw. It was very thrilling.”

Claire Vo · 10:00

“it just hit me with enough joy and a enough utility when it wasn't deleting my calendar that I knew something was there.”

Claire Vo · 10:30
#openclaw#product-market-fit#setup#story
Story52:30

Sam The SDR Agent Replaced 10 Hours A Week Of Sales Work

Claire runs ChatPRD mostly solo, and her sales agent Sam handles the PLG sweep every morning: scanning the CRM for the last 24 hours of sign-ups, identifying company domains and decision-makers, and drafting soft, helpful outreach. It replaced a friend she'd paid 10 hours a week, and it's endlessly tunable by conversation.

  • Sam sweeps the CRM daily for new sign-ups with company domains
  • He uses people-search to find decision-makers, then drafts soft outreach emails
  • He checks with Claire before emailing very large companies as the founder
  • It replaced ~10 hours/week of paid human work and is tuned just by talking to it

“last year before the beginning of the year, I was paying somebody 10 hours a week to do this.”

Claire Vo · 52:30
#sales#automation#use-case#chatprd
Story57:30

Finn, The Family Agent That Runs Household Logistics

Claire's favorite agent, Finn, manages the chaos of three kids across schools and sports. When a basketball tournament schedule arrives, her husband just pastes the page and Finn drops games on the calendar, flags conflicts, and forces the parents to decide logistics. Every afternoon it pings the family chat asking who's picking up whom.

  • The basketball team only reveals game times on Thursday before the weekend
  • Her husband pastes the tournament page and Finn adds games to the calendar
  • Finn flags scheduling conflicts and pushes the parents to resolve them
  • A daily ~3pm ping asks 'which of you is picking up which kids?'

“the basketball team will not tell you when the game is until Thursday before the weekend.”

Claire Vo · 57:30

“every afternoon at about 3:00, it pings in this group chat, and it says, "Which of you are picking up which kids?”

Claire Vo · 59:00
#family#use-case#calendar#logistics

Tool· 3

Tool1:20:30

Ramble Mode: The Highest-Bandwidth Way To Talk To An Agent

A tip from Hillary Gridley that Claire calls the 'Yappers API': the highest-bandwidth way to instruct an LLM is just talking to it. Instead of hunting for APIs or no-code editors, you ramble about what you want, ideally as a phone voice note, and it makes sense of it. She recommends first-time onboarding be done in voice.

  • The highest-bandwidth interface for an LLM is simply chatting to it
  • Ramble about the messy real goal instead of specifying APIs or labels
  • Do it as a voice note on your phone (Telegram uses Whisper transcription)
  • She recommends onboarding a new agent for the first time in voice

“the Yappers API, which is the highest bandwidth API for an LLM is just chatting to it.”

Claire Vo · 1:20:30

“I often tell people when they're onboarding to an open claw, do it on your phone and do it in voice for the first time”

Claire Vo · 1:21:00
#voice#onboarding#tips#openclaw
Tool1:12:30

Run A Headless Mac Mini With Screen Sharing And Remote Login

Instead of dedicating a monitor, keyboard, and mouse to each Mac mini, Claire turns on Screen Sharing so she can pull up the mini's screen from her laptop over the same Wi-Fi. For technical users, Remote Login lets you SSH into the mini's terminal. You still need a monitor and keyboard once, to enable the setting.

  • Enable Screen Sharing in Mac mini settings, then view it from your laptop on the same Wi-Fi
  • No dedicated monitor, keyboard, or mouse needed after setup
  • Remote Login gives SSH access to the mini's terminal for technical users
  • You do need a monitor/keyboard once to turn the settings on

“Go into your Mac mini settings and turn on screen sharing.”

Claire Vo · 1:12:30

“You don't need a monitor, you don't need a keyboard, you don't need anything.”

Claire Vo · 1:13:00
#mac-mini#setup#hardware#tips
Tool1:22:30

Make Claude Code The 'God Mode' Administrator Of Your Agents

Claire's most technical tip: install Claude Code or Codex on the same machine as OpenClaw and use it as the surgeon and manager of your agents. Since OpenClaw is mostly configuration code, Claude Code can read the docs, diagnose a broken email connection, fix the field, or perform 'brain transplants' that split one agent's memory into another.

  • Install Claude Code or Codex on the same machine as your agents
  • OpenClaw is mostly configuration code, which coding agents are good at fixing
  • Point it at the docs to diagnose and repair broken connections
  • It can also 'fracture off' one agent's memory into a new specialized agent

“install Claude Code or Codex on the same computer you're running your open claw on and make Claude Code the god mode administrator of your…”

Claire Vo · 1:22:30
#claude-code#troubleshooting#openclaw#tips

Takeaway· 4

Takeaway10:30

Judge AI Tools By Where They'll Be In A Month, Not Today

Claire Vo explains why she stopped writing off new AI tools on first contact. Because these tools improve so fast, evaluating them on their day-one state is misleading. She came back to OpenClaw (and earlier, Claude Coworker) week after week until she found the unlock.

  • New AI tools ship multiple times a day, so a first impression is a poor signal
  • Spend enough time to see where a tool will be in a week, a month, not just today
  • She nearly wrote off Claude Coworker too, then found the unlock by returning repeatedly
  • Being an 'anti-hype' skeptic is about giving honest, current opinions, not being contrarian

“you really have to pull the thread on these tools and you have to spend enough time with them to see not where they are…”

Claire Vo · 10:30
#ai-tools#adoption#openclaw#skepticism
Takeaway21:00

Never Run OpenClaw On Your Main Computer

Although the install instructions suggest running it locally, Claire argues for a clean, physically separate machine. The agent can do anything a human at your keyboard could, so running it on your work computer risks deleted directories, changed configs, and misdirected files.

  • Assume OpenClaw can do anything a human could do on the machine it runs on
  • Use a clean separate machine: an old laptop, a cloud box, or a Mac mini
  • On a work computer it could delete a directory or send a file to the wrong place
  • Physical separation of the agent's workspace and yours is simply more secure

“would you leave your laptop open and let your assistant run wild on it 24 hours a day? Probably Probably not.”

Claire Vo · 21:00
#openclaw#security#setup#hardware
Takeaway41:00

The Real Unlock: Many Purpose-Built Agents, Not One

Claire says people stumble by throwing every task at a single agent and getting frustrated when it forgets things or loses access. The root cause is context overload. She manages context by sectioning tasks across specialized agents, using a Slack-channels analogy: separate lanes that only intersect when needed.

  • A single overloaded agent forgets conversations and drops tool access
  • The cause is context-window overload, familiar from long ChatGPT or Claude Code sessions
  • She partitions work across agents (Polly for work, Finn for family, etc.)
  • Analogy: nine focused Slack channels, not one 'general' stream everyone must process

“they think they can throw any task at a single agent and get great results. And then they get really frustrated.”

Claire Vo · 41:00

“I have nine Slack channels with my team. And my marketing team's in one, and my sales team's in another”

Claire Vo · 43:30
#openclaw#context-management#agents#workflow
Takeaway55:00

Build Agents That Make The User Feel Like A Winner

Claire's agent Howie sends her a friendly, hyped-up briefing before every podcast recording, complete with who the guest is and a LinkedIn link. The lesson for anyone building agents: it's not just a tool doing work, it's a teammate that makes the user look better to customers and family.

  • Howie sends warm, personalized pre-meeting reminders that hype her up
  • The agents make her look better to customers and show up better to family
  • Cited principle: the best way to win in B2B SaaS is making someone look good
  • For builders: design so the end user feels like a winner

“It is a team helping me look better to customers, helping me honestly show up better to my family”

Claire Vo · 55:00

“thinking about how do I make the end user feel like a winner is a really powerful model to to build a a useful agent.”

Claire Vo · 55:00
#product-design#agents#use-case#b2b-saas