Already Here / Episodes / Episode 3

EP. 03 · AUGUST 6, 2026 · 44 MIN

With Thanh Pham

Thanh Pham is an AI consultant who trains leaders and teams to use AI and helps implement agents inside their companies. He adapts the starting toolkit to each person’s fluency, then looks for one repeated workflow that can prove value quickly. Follow Thanh on X.

Thanh does not begin an AI setup with a long list of tools. He begins with one repeated task, one result a person can recognize, and one reason to use it again. Michael sees that approach applied to automatic meeting briefings, lead enrichment, separate responsibilities for different assistants, and local versus cloud setups. Thanh’s 10-80-10 model keeps the person responsible for defining the job and reviewing the final result.

Watch on YouTube · Listen on Spotify · Listen on Apple Podcasts

What you’ll learn

Learn why adoption starts with the executive’s current fluency and daily work. Thanh gives one client simple document and report tasks while another works in the terminal and wants advanced strategies; the useful next step matters more than the most impressive stack.

Learn how one useful result on a regular cadence builds trust and earns the next workflow. Thanh leaves each newly onboarded agent with one strong skill that can produce visible value before more capability is added.

Learn the practical tradeoff between a tangible local agent and cloud-based model capability. A machine such as a Mac mini can make files, skills, and ownership more visible, even when the language model itself still runs in the cloud.

Learn why clear roles work better than one agent doing everything. Thanh’s OpenClaw setup handles email, calendar, and meeting briefings, while Hermes takes on research, prototyping, coding, and project work.

Learn how a pre-meeting brief can combine messages, email history, and public research into useful context. The workflow proves value by preparing work an executive already needs, without demanding a large change in behavior.

Learn how lead enrichment becomes valuable when it helps a team decide fit faster against its own criteria. Public details such as industry, background, and profession turn a slow, repeated qualification task into a decision the business can make in seconds.

Learn Thanh’s 10-80-10 model: people define the outcome, resources, access, and definition of done; the agent handles the middle; and a person reviews the final result. Most of the production can be delegated without delegating the judgment.

Learn why a meaningful personal project teaches more than passively watching tutorials. Thanh recommends building your own version even when an app already exists, because solving a problem you care about creates a reason to understand how the system works.

Topics and resources

  • OpenClaw — the personal AI platform Thanh uses for assistant work
  • Hermes Agent — the generalist agent in Thanh’s research and project workflow
  • ChatGPT — one of the cloud AI products discussed in the episode
  • Claude — used in the workflows Thanh describes
  • OpenAI Codex — used as a reviewer in Thanh’s 10-80-10 workflow
  • Mac mini — the local machine used in several personal-agent setups

One safe experiment

Choose one repeated task and write down the result you need, the information the assistant may use, and what “done” means. Ask the assistant to prepare one output, inspect it yourself, and keep the final action or decision on your side of the line. Add more capability only after that first result is useful and easy to review.

Browse every episode · Previous: Matt Van Horn · YouTube · Spotify · Apple Podcasts