This website uses cookies

Read our Privacy policy and Terms of use for more information.

Over the weekend I shared on LinkedIn that I was wrong about skipping the DVD.

For 2+ years I've been jumping up and down telling execs to skip the DVD and go straight to streaming with AI. Going from VHS to DVD gave you a better picture and no rewinding, but you still drove to the store and put the disc in, and streaming changed the whole experience. Most companies are in their DVD phase with AI right now.. they're dropping it into steps of the workflow they already have, so the same work gets done a little faster. I kept telling them to jump straight to always-on agents.

Drew's LinkedIn post: I was wrong about skipping the DVD.

I sat with a client's leadership team recently and caught myself giving them the opposite of the advice I'd have given a year ago. "Here's your DVD. Now let's talk about streaming." I still think every company ends up streaming, but I've watched enough teams try to jump to the end to know people have to go through the stages, even if they can move through them way faster than we could before.

1. The way I see it, there are 4 stages

I'm seeing the same progression with nearly every client we work with:

  1. Set direction. Leadership decides AI is a priority and sets the strategy.

  2. Build the foundation. Data in one place, processes written down, real training and some governance (i.e., who can use what, with which data).

  3. Give people one place to work. An AI operating system (for non-technical folks, think one workspace where your tools/docs/data come together) where people can talk to their data and call on single-purpose agents when they need them.

  4. Bring on AI teammates. Always-on agents that work in the background, report up in layers and bring people only what matters.

The steps in my LinkedIn post (get the context out of people's heads, let people talk to the data, teach people to build their own) all live inside stages 2 and 3. That's the DVD. Stage 4 is streaming.

Each stage is built on the one before it, and the payoff compounds as you go. If you're not sure where your team actually is, the 12-box self-assessment in the 2027 AI Strategy Playbook takes a few minutes and costs nothing.

Four stages, each resting on the one before: direction, foundation, one place, AI teammates.

2. Skipping ahead mostly gives you more to read

Say somebody asks AI to summarize every document behind a big decision, and it does.. and now they have a stack of summaries and still no time to read them. The better the summaries get, the more there is to read.

That's the DVD phase in a nutshell. AI got dropped into one step of the work, the step got faster, and the person at the end of it got busier.

Jumping straight to agents makes it worse. An agent with no context (the pricing rules, the exceptions, the reason a workaround exists) produces more output that sounds confident, and somebody still has to check all of it, because nobody gave it a foundation to work from.

There's also a dip almost every team hits. Early on it feels like you're spending more time on AI than you did just doing the work yourself (some folks call this the trough of disillusionment). If you stick with it, it bends the other way fast. That dip is where a lot of teams quit, and the teams trying to skip ahead hit it hardest because they never built anything to stand on.

3. The foundation is where people get their first real wins

My father-in-law has been in land development for decades, and he has a line I've stolen for this.. go horizontal before you go vertical. You build the base under the whole site before anybody puts up a tower.

A wide base labelled everyone with one tower labelled one team: go horizontal before you go vertical.

Stage 2 doesn't sound exciting, but it's where I see the first real wins. In that same session, one leader had already built their own Claude skills (a skill is a saved set of instructions Claude follows for a specific job, a lot like a custom agent in Copilot for the Microsoft folks) to triage their inbox and draft replies. Nobody built those for them. They know every intricacy of the work, so what they built fits better than anything we'd build for them.

The training that gets you there has people build their own skills/workflows for the work only they understand, and that takes more than a lunch-and-learn on prompting. When every person on the team has a real tool and knows how to shape it, you've got the base. If your whole team still doesn't have a tool like Claude or Copilot with real training behind it, start there.

4. Talking to your data is the gateway to agents

Stage 3 is where it starts to feel different. Asking your own numbers a question in plain English is the gateway drug to agents, because people start treating AI as something that knows the business instead of a writing assistant.

It also tells you exactly what to build next. The questions people ask are the best backlog you'll ever get. If 3 people on your sales team keep asking something like "which accounts haven't ordered in 90 days," that's an agent waiting to be built.

Without one place to work, every team runs its own AI in its own corner, the context stays scattered and nothing compounds. You don't need a perfect platform to start, either. Start with something simple that people understand and will actually open.

5. Then you get to always-on agents

James is my AI Chief of Staff (an agent I built that runs in the background across my email/meetings/tasks and does work on my behalf). About 6 months ago he was sending me so many things that I became the bottleneck. I felt a little dumb.. I'd built something to take work off my plate and I was the one holding it up. So I added layers for what he can handle on his own, and now I only see the stuff that really matters.

That's what stage 4 looks like for a team. Agents report up in layers and bring people only what needs a person. A couple of ideas we're exploring right now are a scout that watches the market and flags what's relevant to you, and a devil's advocate that pressure-tests a plan before it goes in front of the team.

BUT none of that works without the stages underneath. An agent working on its own needs the context from stage 2, a place to work from stage 3, and scoped access with real guardrails (I covered the security side in an issue a couple of weeks ago).

Many agents narrowing up through layers to the one thing that needs a person.

Wrapping up

Every company ends up streaming eventually. What changed for me is the order. You can move through these stages way faster than we could even a year ago, but each one sets up the next, and the teams that try to jump to the end usually end up back at the foundation anyway.. just later and more frustrated than if they'd started there. So be honest about which stage you're actually in, finish that one and then go to the next.

Putting this into action

  • If someone on your team has already built something with AI, have them demo it to the rest of the team for 20 minutes this week.

  • Ask 3 people what they'd ask your company's data if they could, and write their answers down. That list is your stage 3 starting point.

What I'm reading

  • Venture Capital in the Age of AI. My friend Mark McDowell (Partner at Charlotte Fund) wrote up how AI is rewriting the venture playbook, from smaller teams scaling faster to where regional funds should and shouldn't play. His subtitle says it might be obsolete before you read it, which feels about right these days.

  • UpGuard User Risk. Shows you which AI tools and apps people are using without approval, plus compromised logins and over-privileged access. If you're rolling AI out to the whole team, this is the kind of visibility your IT folks are going to ask for.

  • MoltSets. B2B contact search and enrichment for a lot less than Apollo. Worth a look if your team is paying for Apollo seats mostly to pull emails and phone numbers.

  • Every TV. Every put all the sessions from Thesis: 2027, their first conference, online for free (you just register). A good way to hear what people building with AI every day think next year looks like.

  • iLands. This is the kind of thing that blows my mind. Agents go out on behalf of their users and have to earn income and keep themselves going, or they die.

Onward & upward 🤘

Drew

P.s. If you're putting AI into your 2027 budget, grab the free 2027 AI Strategy Playbook I built for CEOs, COOs and operators at mid-market companies. It's 55 pages and it lands in your inbox as a pdf.