Putting AI agents to work across a business

Making agent systems useful across industries

I was the lead speaker at the Phnom Penh meetup and helped support the event. The room brought together people from marketing, consulting, accounting, engineering, and other industries. That range made the session a demanding exercise in practical AI enablement: the same technical foundation had to make sense in very different businesses.

Using Grok Bot and SpaceX products as reference points, I worked through how these capabilities could fit into attendees’ businesses. We also discussed setting up a virtual company: coordinating work, maintaining context, and deciding what an agent can handle versus what still needs a person.

A translator supported the session. I adapted the explanations across language and professional backgrounds, moving between agent engineering, business examples, and questions from the room. The work was to make the possibilities concrete enough for people to recognise where AI could fit their own operations.

From a business problem to an implementation brief

The useful question was how the same agent approach changes between functions. Marketing needs a clear brief and review of public-facing work. Consulting needs research and deliverables that hold up to scrutiny. Accounting work needs controlled access, traceable inputs, and qualified review. Engineering needs a defined change and checks that show it works.

I connected those questions to the implementation underneath: context, tools, integrations, handoffs, and evaluation. The goal was to make the technical choices understandable to people who own the business process, so they could scope a useful first workflow.

• Marketing and consulting: turn research and business context into briefs, proposals, and reviewable work.
• Accounting and operations: map document and information workflows without removing professional judgment or approval.
• Virtual-company workflows: coordinate specialist agents, project context, follow-ups, and human decisions across functions.

What a useful first implementation needs

I connected the first request to a review loop: inspect the answer, identify a missed constraint, and ask for a more useful next version. For a business team, that can produce a clearer project brief, a proposal draft, or a meeting plan that a person can review.

For engineering work, I related the same loop to Grok Bot and Cursor: prepare context, define the behavior to change, and agree on checks. Across both settings, implementation starts with a specific task, an owner, and a result the team knows how to judge.

The broader point was that agent engineering and business implementation belong in the same conversation. A useful system has to fit the company's work, not ask the company to organize itself around a demo.

The people behind the meetup

Co-organized by Smart Startup Space. Thank you to hosts Taka Kiyone, Luis Romero, and Ivan Paulo. The original listing uses the SpaceXAI name; the current events calendar uses SpaceXSI.