Building the company: team, AI-assisted operations and scaling

0:00 Building the company

At some point, every founder hits the same wall. You're working every hour, you're the bottleneck for every decision, and the business can't grow any faster than you can personally move. That's the moment you have to stop being just a founder and become a company builder. In this lecture, you'll learn how to build an early team, set an operating rhythm, use OKRs, design AI-assisted operations with humans in the right places, and look after yourself while you do it.

0:35 Why it matters

Why does this matter? Because the habits you set with five people become the culture of fifty. If decisions only happen when the founder is in the room, the company stalls. If nobody writes anything down, every new hire relearns the same lessons. And in 2026, there's a new twist. Small teams can do far more with AI assistance, but only if the work is designed well. Here's the key idea. Build systems, not heroics. The goal is a company that works when you're not in the room.

1:13 The kitchen analogy

Here's an analogy. Think of a professional kitchen. A home cook can make a great meal alone, improvising as they go. A restaurant can't work like that. It has a head chef, stations, recipes written down, a rhythm for service and prep, and quality checks before a plate leaves the pass. Modern kitchens also use machines for repetitive work: mixers, slicers, combination ovens. But no restaurant lets a machine decide whether a dish is good enough to serve. That's AI-assisted operations in one picture. Machines for the repetitive, well-defined steps. Humans at the pass.

1:54 Team and rhythm

First, the team. Co-founders should bring complementary skills, shared values and clear roles, with a written agreement. Early hires should be adaptable people who are comfortable with ambiguity and will build processes, not just follow them. Hire for the next twelve months, not the next five years. Next, the operating rhythm. A short daily check-in, in person or asynchronously in Slack or Teams. A weekly metrics review. A monthly look at forecast versus actuals and runway. And quarterly goals. Then OKRs, objectives and key results: a few ambitious objectives, each with measurable key results, reviewed regularly.

2:36 AI-assisted operations

Now AI-assisted operations. Good candidates are high-volume, well-defined, reviewable tasks: summarising calls, drafting first versions of documents and emails, triaging support tickets, extracting data from invoices and preparing reports. Poor candidates are high-stakes judgements, anything you can't delegate accountability for, and tasks where errors are hard to spot. A quick test for any task: how often does it happen, can you define what good looks like, what does an error cost, and who checks? Then the principles. Write the SOP first. Put a human in the loop where errors matter. Measure quality and time saved. Protect data. And train the team.

3:20 Worked example 1: call summaries (illustrative)

A simple worked example, illustrative. Ella runs a two-person bookkeeping practice in Glasgow. Every client call ends with her typing notes into the CRM, about fifteen minutes each. She tries an AI meeting summary tool on a business plan with appropriate data settings, after getting clients' consent to record. The tool drafts the summary and action items. Ella skims and corrects each one, which takes about three minutes. She samples five a week for accuracy. Over a month, she saves several hours and her notes are more consistent. But she doesn't let the tool send anything to clients. She's the pass.

4:04 Worked example 2: a Karachi agency (illustrative)

Now a realistic scenario, illustrative. A six-person digital agency in Karachi is drowning in admin. The founder, Asad, lists every recurring task for two weeks. He writes SOPs for the ten most common. Then he picks three for AI assistance: call summaries into the CRM, first drafts of monthly client reports, and extracting invoice data. Each has a named human reviewer and a weekly quality sample of ten outputs. After two months, admin time is noticeably lower, client-facing quality is owned by people, and he uses the time saved for sales rather than hiring another coordinator straight away.

4:47 Watch me: an SOP with an AI step

Watch me write an SOP with an AI step, using the template in the lesson. Name: monthly client report. Owner: account manager. Purpose and definition of done: a two-page report with the five agreed metrics, three insights and next month's actions, sent by the fifth working day. Trigger: the first of the month. Step one, export data from analytics. Step two, the AI step: the approved tool, a link to the saved prompt, and the inputs allowed, which exclude client personal data. Step three, human review against a checklist. Step four, send. Then quality: sample five reports a month. And escalation: if metrics drop more than twenty percent, the founder reviews before sending.

5:36 Document how the company works

One more system worth building early: how you document and share knowledge. Keep a single home for how the company works, a wiki in Notion, Confluence or a shared drive, with SOPs, decisions, policies and onboarding guides. Record important decisions in a short decision log: what we decided, why, what we considered, and when we'll review it. And if you have investors or advisers, send them a short monthly update with highlights, lowlights, metrics, cash and specific asks. It keeps them helpful rather than anxious. These habits feel like overhead at five people. At twenty, they're the difference between a team that moves on its own and a team that waits for you.

6:25 Founder and culture

Now, you. Founder wellbeing isn't a luxury. It's an operational risk. Exhausted founders make worse decisions. Protect sleep and exercise, take real breaks, and build a support network of peers, mentors and friends outside the company. Also protect your reputation and ethics. Treat customers, employees and partners fairly, keep promises, and be honest about what your product, including its AI, can and can't do. As you scale, expect roles to change. Communication becomes harder, processes need formalising, and you'll hire people who are better than you at specific jobs. That's success, not a threat.

7:06 Common mistakes

Let's list the common mistakes. Hiring friends for roles they're not suited to. No written roles or decision rights. Meetings with no rhythm or purpose. Too many OKRs. Automating a messy process with AI before it's been written down, which just produces mess faster. Letting AI output reach customers without a human check. Using unapproved tools with sensitive data. And founders who never rest. Finally, know when to pivot or stop. If evidence keeps contradicting your core assumptions, changing course is a sign of good judgement, not failure.

7:44 Recap and try this now

Let's recap. Build systems, not heroics: a clear team, a daily-to-quarterly rhythm, and a few OKRs. Use AI for high-volume, well-defined, reviewable work, write the SOP first, keep humans at the pass where errors matter, measure quality, protect data and train the team. Look after yourself and your ethics. Your try this now: log your recurring tasks for one week, write SOPs for the top three, and pick one for AI assistance with a named reviewer and a weekly quality sample. That completes the course. Now take the final assessment and put it all into practice.