As shared in a reflection on LinkedIn, we’d have laughed three years ago if I had told you we would build a profitable company impacting 300,000+ people with a team of five humans and 54 AI agents. Yet here we are. Now, we’ve received tons of emails asking: How?
The answer isn’t in ‘adopting AI’ like another software update. It’s more straightforward—and far more human. Think of AI not as tools to install, but as apprentices to train.
AI isn’t software you install—it’s an apprentice you mentor. When a new employee joins your team, you don’t “install” them. You orient them, provide guidance, evaluate their work, give feedback, and help them grow.
Treat AI as an apprentice; it is a shift that changes everything, and you will see tangible results.
📊 The Great Mental Shift: Software vs. Apprentices
Microsoft turned 50 last week! While delivering a seminar on leading human-machine teams, I stopped by Building 9 at the main campus to snap the photo shared above and reflect on the incredible 50-year journey of the tech world.
For 50 years, we’ve treated technology as something to install—like Office, Windows, or one of the 100 apps in our phones. Click, configure, control. But AI doesn’t “install.” It onboards.
At AI4SP, we run a global operation with 54 AI teammates and just 7 humans. The breakthrough wasn’t deploying tools—it was learning to:
- Write AI job descriptions (yes, seriously)
- Define “culture fit” for algorithms (hint: it’s not ping-pong tables)
- Measure ROI beyond productivity (innovation + balance > output)
The companies winning this era—like Perplexity, Cursor, Sobek.ai, CustomGPT.ai, Midjourney, ElevenLabs, Anthropic, and Praktika.ai—aren’t just using AI. They’re leading hybrid human-machine teams with entirely new playbooks.
| Traditional Software Approach | Apprenticeship Approach |
|---|---|
| One-time setup | Ongoing relationship building |
| Static capabilities | Continuous learning and growth |
| Operates on commands | Responds to conversation |
| Focus on features | Focus on development |
Our research shows why the apprenticeship model wins:
🔹 Only 10% of info workers master prompt engineering—because they’re stuck “configuring software” vs. *teaching an apprentice*. (Tech vendors obsess over features, not paradigms.)
🔹 Apprenticeship slashes proficiency time by 50%:
- 14 weeks to master general AI tools (prompt engineering)
- 5 days for guided-experience apps (apprenticeship UX)
🔹 Satisfaction gaps don’t lie:
- 80% for single-purpose AI (no prompts, guided flows)
- 60% for general enterprise AI (endless tweaking)
The verdict? Stop configuring. Start coaching.
The apprenticeship model works because of two things
- Providing context (what prompt engineers call “effective prompting”)
- Knowledge transfer (what AI developers call “training data”)
Most AI creators built user interfaces that allow us to treat AI as an apprentice; you can naturally provide context and knowledge without needing technical expertise.
📊 Onboarding AI has tangible benefits
How professionals engage with AI, reinvest the time they save, and how it impacts their perceived work quality.
How they use AI tools
| Level of Use | Description of AI | % |
|---|---|---|
| I do not use AI | I do not trust AI or find it helpful in my work. | 8% |
| AI as a Helpful Assistant | I use AI occasionally as a tool or assistant to simplify tasks and make parts of my job easier. | 41% |
| AI as a Tool | I use AI as a software tool—a creative partner that brings its skills and insights to enhance my work. | 37% |
| AI as My Extended Team | I approach AI as a group of specialized team members.[*] | 14% |
[*] This group typically uses 5-10 tools, treats each as an onboarded teammate, gives regular feedback, and guides their evolution and alignment with goals.
Our research shows the most successful AI users are:
- 25-35-year-olds who treat AI as a collaborative partner.
- 18-25 and 35-45 year olds with similar approaches.
- A significant drop-off in satisfaction and reported productivity increases in the 55+ age group who still use command-based interactions.
Time Reinvestment Among AI Users
| Level of Use | Generating New Ideas | Doing Additional Work |
|---|---|---|
| AI as a Helpful Assistant | 37% | 12% |
| AI as a Tool | 42% | 15% |
| AI as My Extended Team | 58% | 17% |
Perceived Improvement in Quality of Work
| Level of Use | % Agree Work Quality Improved |
|---|---|
| AI as a Helpful Assistant | 52% |
| AI as a Tool | 65% |
| AI as My Extended Team | 90% |
🧠 As part of this research, we evaluated how changing just one thing about how people approach AI tools can dramatically improve results, without changing the tools themselves. 🎧 Check out our podcast for the whole story and discover why this simple shift could be the missing piece in your AI strategy.
💡 The Natural Advantage: Why Frontline Workers Get It
Our October 2024 research showed an interesting pattern: eight out of ten frontline workers can extract needed insights from AI on their first try, compared to just three out of ten knowledge workers.
Why this massive gap? Communication style:
- Frontline workers use prompts averaging 28 words.
- Executives use prompts averaging 5 words.
Frontline workers “chat” with a co-worker, while knowledge workers and executives “command software to do something. ” Hence, frontline workers are more successful.
Frontline workers intuitively treat AI as a teammate they’re guiding, not a machine they’re commanding.
When Teresa, a convenience store clerk, texted our AI agent during a power outage, she didn’t write “food safety rules.” She wrote: “Hey, we have a blackout, there is no power and I don’t know what to do with the food, should I throw it all away?” – actually she wrote it in spanish and the phrase she used was “se fue la luz” which is common in some spanish peaking countries to say the lights went out instead of the power went out.
Teresa interacted as if she were reaching out to friends on social media or to coworkers because, for her, having a conversation was natural.
On our end, AI4SP treated the AI agent (a RAG model) as an apprentice, providing it with knowledge (600 pages of FDA retail food regulations), training it on what persona to adopt, and teaching it how to provide guidance. We onboarded the AI like a new team member, and Teresa used it accordingly.
🔄 The Apprentice Onboarding Process
When I analyzed our most effective AI agents at AI4SP, a number surprised me: the five agents I use most- our COO, our lead researchers, speechwriter, product manager, and meeting coordinator—have been trained with approximately 30 million words. That’s almost double what a PhD candidate reads during their entire undergrad and doctoral program, including dissertation research.
These words aren’t abstract data. They’re the books that changed my thinking, the emails that shaped deals, the original research we pioneered. They’re every speech I’ve given, every consulting engagement that challenged me, every late-night discussion with CEOs about AI’s future.
“I read 50 books a year—their ideas shape me, but their details slip away. My AI apprentices never forget. They carry every lesson forward, turning my accumulated experience into an always-available advantage.”
Step 1: Orientation (Not Installation)
Assign purposeful work—Start with a single task in your domain of expertise, and clarify goals.
Step 2: Knowledge Transfer (Not Configuration)
Treat it like a teammate—Share documents, examples, and feedback to shape its understanding.
Step 3: Ongoing Development (Not Maintenance)
Grow its role—Review performance, expand responsibilities, and keep knowledge current.
Practical Tip: Include your AI agents in all relevant communications, just as you would with employees. This makes the learning process continuous without taxing your time.
🚀 Building Your First AI Apprentice: The Personal Coach Example
Let’s make this concrete with a simple example anyone can implement today:
- Choose your focus area: If you’re a runner who tracks performance stats, this is perfect starting material
- Gather your knowledge: Collect your running logs, preferred training approaches, and goals
- Create a private project: Use the “projects” functionality of ChatGPT or Claude, with privacy settings enabled
- Begin the conversation: “I’d like to create a running coach to help me improve. Here’s my training data…”
- Provide feedback: After initial responses, guide the AI on what’s helpful and what needs improvement
Within a week, you’ll have a personalized coach understanding your patterns and preferences. More importantly, you’ll understand how to build AI apprentices for other areas of your life and work.
AI4SP inside info: When we create our most sophisticated agents and assistants, we ensure we can switch between models: GPT, Claude, Llama, Mistral, Gemini, Yi, etc, as they are commodities. The value is in our knowledge curation. Similarly, we favor AI tools that allow us to choose which LLM we want to use.
🔮 One More Thing…
Tonight’s challenge:
1️⃣ Pick your most repetitive task.
2️⃣ Onboard an AI apprentice to own it.
By this time next week? You won’t be a solo worker—you’ll be a leader of a hybrid team.
The future isn’t built with better tools.
It’s built with better teams—where the rarest skill isn’t prompt engineering, but apprentice development.
📚 Resources
- Digital Skills Compass: Free assessment in 7 languages at skills.ai4sp.org
- Workshops & Training: Book sessions for your team
- Complete Research: Request our detailed findings
Luis J. Salazar
Founder | AI4SP
Sources:
Our insights are based on +250 million data points from individuals and organizations who used our AI-powered tools, participated in our panels and research sessions, or attended our workshops and keynotes.
