Last year, I stumbled into AI consulting.
A friend asked if I could help his private equity firm build an AI product. Since then, I've helped Fortune 100 companies, professional services firms, and several startups build and leverage AI.
Here are a handful of lessons I've learned:
1. Leaders aren't using AI
Most leaders I meet at large companies are not using AI. Every week, I see leaders surprised by basic features like speech-to-text or Shift+Enter in the chat box.
They're asking their employees to adopt AI, but they aren't leading by example. This makes it harder to drive adoption and realize ROI.
They're also not developing an intuition for what AI can do. Without firsthand experience, it's hard to build a vision for what's possible and how it will impact their company.
My friend Geoff Woods has turned the AI lights on for more leaders than anyone I know. He has a simple formula for getting started: Once a day, every day, for the next 30 days, ask yourself, "How can AI help me do this?"
Lesson: Use AI ambitiously and lead by example.
2. AI slop is a big problem
The amount of AI slop shared within and across businesses continues to amaze me.
People (meat proxies) are outsourcing their thinking to AI and passing the results off as work (slop grenades).
Nicholas Charriere summarized this really well:
before, when your coworker didn't care you got no output.
now, when your coworker doesn't care you get slop.
imo it's worse
AI output without thinking creates more work for others, not value.
Lesson: As AI makes output cheaper, thinking and brevity become more valuable. Do not tolerate slop.
3. Workflow discovery is the real bottleneck
Every company has workflows where AI can make an enormous impact. The hard part is identifying those workflows, prioritizing them, mapping how the work actually gets done, and training employees on a new way of working.
This is so hard.
The people closest to the work are usually too busy to step back and map it, and most teams don't have the skill set to do it well. How people say the work gets done is not always how it actually gets done. And once there's a new way of doing work, people are resistant to behavior change.
For many AI startups, this is literally the whole opportunity. You either own a workflow or a set of workflows for customers, or you compete with those customers by automating the workflows faster.
Lesson: Understand how the work actually gets done. Prioritize the most impactful workflows and automate pieces for quick wins.
4. Your customers think they can build your product
If you're selling software to a company with an engineering team, they probably think they can build what you're selling, and they're actively exploring how to do it. Tools like Claude Code, Lovable, and Replit have given companies the confidence to think they can build everything.
I believe this mindset is ridiculous and temporary. Just because it's easier to build software doesn't mean companies should build everything. Is it strategic for your company to build, maintain, and improve a meeting notetaker or a CRM from scratch, or should you use a highly crafted solution from a partner?
That said, I do think this mindset will compress prices and make it harder to sell crappy software at ridiculously high prices. a16z shared a Build vs. Buy graph that illustrates this beautifully.
Lesson: Craft products so valuable that buying them is an easier decision than building them.
5. Products enabled by people will win
The most impactful AI solutions will likely require customers to change how they work. A product without service to help them make that change can fall short on value. Service without product can be inefficient and costly.
The winning playbook feels like a combination of both: AI-powered products enabled by service, or service empowered by AI-powered products. You're not just selling a product. You're selling outcomes delivered via product, made possible by people.
Palantir pioneered this. They built a suite of powerful products that are embedded into organizations and continuously improved by forward-deployed engineers (FDEs). But FDEs aren't the only playbook. I think we're going to see a lot more models like this.
Lesson: Take on more of the solution if it helps your customer achieve their desired outcome.
6. Professional services have the most to gain or lose
AI is an existential threat to professional services firms in law, accounting, and insurance, but also an enormous opportunity. They have what every AI startup wishes it had: a deep understanding of the workflows, boatloads of data, and strong customer trust.
All they have to do is add AI, right? Unfortunately, it's not that simple. Legacy business models (hourly billing), partnership structures, lack of AI talent, and resistance to change make it incredibly difficult for professional services firms to adapt.
Lesson: For established firms, rethink how you deliver services, not just how you use AI. For startups, use this window to compete with incumbents or own a workflow and make them your distribution partners.
7. Trust matters more as AI gets noisier
Knowing what's good and real in AI is getting harder. It's easier than ever to build. There's an explosion of new things, which is awesome, but there's also a ton of noise and a lot of snake oil salesmen.
Most people aren't working in AI or spending all day on X. For them, it's really hard to know what to trust. This makes it tough to know which AI products to adopt or companies to join.
I've often felt that the most valuable thing I do is use and consume as much AI stuff as possible, try to make sense of what's happening, and act as a filter for people.
Lesson: Build trust, invest in your brand and partnerships, and play the long game.