Houston, we have a problem: More than 80% of enterprises report no tangible impact on EBIT from their generative AI investments, according to McKinsey’s March 2025 research. Meanwhile, S&P Global’s 2025 report shows that 42% of enterprise AI projects are now abandoned before reaching production, a significant increase from just 17% the previous year. Our AI4SP Global tracker shows 80% satisfaction with off-the-shelf AI tools and less than 40% satisfaction with enterprise-sanctioned AI deployments.
But here’s the kicker: While corporate IT departments struggle with billion-dollar AI initiatives, ChatGPT has exploded to 800 million weekly active users, driven entirely by grassroots adoption. The contrast is stark and telling.
Why?
The grassroots are speaking. OpenAI reports 3 million paying business customers, who often arrive after first experimenting with the public ChatGPT site, with IT departments later purchasing enterprise licenses to regain control. This bottom-up demand drove OpenAI’s annual recurring revenue to $10 billion. On a more modest scale, AI entrepreneurs, from Base44 (acquired for $88M by Wix weeks ago after just 6 months of operation), to Windsurf, Cursor, and Perplexity AI, reaching multibillion-dollar valuations, share a familiar tale we observe worldwide:
AI adoption succeeds when it starts with the people closest to the work, not the boardroom; and tech providers optimizing for grassroots AI adoption are winning the first leg of the race.
Learning by Doing (Not Waiting)
Whether it’s data from 85,000 individuals taking our AI Compass assessment, or lessons from Fortune 500s, universities, and governments, the winning formula is always the same: start at the grassroots.
Gartner predicts that over 40% of agentic AI projects will be canceled by 2027 due to escalating costs, unclear business value, and inadequate risk controls. The same research shows that by 2028, 15% of day-to-day work decisions will be made autonomously through agentic AI, but only for organizations that first master the fundamentals.
Mastering the fundamentals happens bottom up!
🎧 Listen to bonus content on the podcast version of this article: Apple, Spotify.
The first wave of AI value isn’t about moonshots—it’s about automating friction that’s slowed teams down for years. The magic isn’t in a million-dollar project plan. It’s the frontline worker who spots a bottleneck, writes a prompt, and saves their team an hour a day.
That’s how Elizabeth—our AI COO—was born. She started as a simple prompt to rewrite a LinkedIn post. Now, she’s running operations, managing 20 million words of knowledge, and delivering the output of 10-12 people for the cost of a single mid-level hire. The ROI? Fifty times. And it started with a $20 experiment 20 months ago.
Writer’s March 2025 research shows that enterprises without a formal AI strategy report only 37% success in AI adoption, compared to 80% for those with a plan, but the approach must enable grassroots experimentation, not stifle it.
🛣️ Two Paths, Two Outcomes
Grassroots MVP Approach:
Teams start with prompts, build knowledge, create personas, automate workflows, and evolve into agentic AI. Each stage delivers real value—fast.
Top-Down Approach:
Big budgets, big teams, endless meetings, and a year-long wait for results. By the time the “perfect” agentic AI launches, the world—and your business—has moved on.
“Grassroots AI adoption succeeds 70% of the time. Top-down or IT-driven initiatives? Less than 20%. But when we intervene in our strategic engagements, evolving top-down initiatives into “guided” grassroots momentum, success jumps to 90%. We’ve seen it across 100 organizations.”
Time to Value: Grassroots vs. Top-Down
| Deployment Model | Time to First ROI | Productivity Lift (6 Months) | Success Rate |
|---|---|---|---|
| Grassroots MVP | 2 weeks | 2–3x | 70% |
| Top-Down/IT-Driven | 6–12 months | 1.1–1.4x | <20% |
| Hybrid (Top-Down Guided Grassroots) | 1–2 months | 2x+ (compliant and secure) | 90% |
Source: AI4SP Global Tracker, Jul 2025. +115,000 organizations and individuals from 70 countries
🪜 The Five Stages to Agentic AI
- Prompting: Quick wins with smart prompts—get value on day one.
- Knowledge and Context Curation: Feed your AI curated, high-value information. Proprietary knowledge is your edge.
- Persona and Context: Define your AI’s job description, boundaries, and tone. Treat it like onboarding a new hire.
- Workflow Automation: Integrate AI into daily processes—emails, Slack, meetings, you name it.
- Agentic AI: Achieve autonomy and compounding productivity. Your AI becomes a true teammate.
🤖 The Human Shift: From Doer to Director And the Unexpected Management Revolution
Our research across 100+ organizations shows a fascinating behavioral shift: individual contributors are becoming managers without realizing it, fundamentally rewiring organizational dynamics.
As teams adopt AI agents, someone naturally emerges as the “agent manager”—typically the person who started the experiment. But unlike traditional management hierarchies, these new managers aren’t constrained by HR privacy policies, performance review cycles, or territorial ego battles. The result is Management transparency and collective learning.
Management is still needed; the notion that AI agents are “set and forget” is just a myth. The best results are achieved when we manage our agents like apprentices, providing daily feedback, knowledge updates, and prompt adjustments.
Context engineering is a critical skill in AI; it involves providing AI with the necessary data to make informed decisions. This isn’t a technical function. Context is how your company operates: the ideal versions of your reports, documents, processes, and organizational tone. Don’t just dump every document into a search system (RAG) and hope for the best. Frontline experts (grassroots) must make deliberate choices about what context matters, build ideal versions of key documents, and treat context engineering as the cross-functional challenge it is.
At AI4SP, we have three people managing Elizabeth, our AI COO. However, she delivers output equivalent to 12 people. ➡️ The math isn’t 3:1; it’s 3:1:12.
So, watch out for finding your ideal balance and understand that the organizational charts of companies with AI agents show unorthodox spans of control ratios.
Teams are running “co-management standups”—15-minute sessions to debug, share tips, and collectively improve their AI agents. There are no privacy issues, no HR rules or policies, just pure learning.
Why this matters for grassroots adoption: Traditional management structures create bottlenecks. When a director manages 8 people, information flows slowly. However, when three people co-manage an AI agent delivering 12 times the output, knowledge transfers instantly. Problems get debugged in real-time. Best practices spread organically.
A positive change: Teams with AI agents spend 60% more time on collaborative problem-solving and 40% less time on status updates. They’re managing outcomes, not people. And they’re doing it better than traditional hierarchies ever could.
This is a fundamental reimagining of how work gets organized. And it’s happening bottom-up, driven by teams who refuse to wait for permission to get better results.
⚠️ The Governance Gap: Why Policies Are Failing Teams
Only 18% of organizations have proper AI governance councils, yet governance is one of the most significant predictors of AI success. Governance councils move slowly, and innovation is moving very fast; as a result, policies written for GPT-3.5-era models are blocking teams from using Claude Sonnet 4 or GPT-o3—technologies that are fundamentally different in capability and risk profile. Teams are bypassing these outdated policies and instead using shadow AI.
The grassroots teams aren’t waiting. They’re finding workarounds, using personal accounts, or worse: abandoning enterprise-wide AI initiatives entirely, in favor of shadow AI. The governance intended to protect organizations is preventing them from capturing the value of AI.
✅ Three Things You Can Do Now
- Experiment Relentlessly:
Start with prompts and simple automations. Use synthetic data if necessary. Learn by doing—don’t wait for the “perfect” use case. - Modernize Your Policies:
Review and update your data and governance policies. Remove blockers. Make sure your rules are built for today’s AI, not yesterday’s. - Change Your Mindset:
Managing AI is like managing apprentices. Block 15 minutes a day for agent management—solo or as a team. Share learnings, debug together, and treat it as a core leadership skill.
Watch for sycophancy: We’re starting to worry about this more than hallucinations. It’s when AI abandons its correct assumptions just because you say the opposite. This is not the obvious “you’re so brilliant!” response, but the subtle agreement that undermines good decision-making. Elizabeth’s value as our AI COO would quickly diminish if she always agreed with me. We’ll dive deeper into this challenge in a future article.
🔮 One More Thing
Jeff Raikes taught me that “We overestimate our impact in the short term and underestimate the long-term consequences of our actions.” Elizabeth started as a $20 LinkedIn post rewriter. Twenty months later, she’s delivering over a million dollars worth of value. That’s the power of starting small and thinking long-term.
Start with a simple prompt today, and let compound learning build your competitive advantage.
🚀 Ready to Take Action?
- Share this article with a colleague or educator
- Workshops & Training: Book sessions for your team
- Complete Research: Request our detailed findings
✅ Ready to transition from a traditional organization to an AI-powered one?
We advise forward-thinking organizations to develop strategic frameworks for evaluating, integrating, and optimizing human-AI production units. Contact us to explore how we can support your organization’s evolution in this new talent landscape.
Luis J. Salazar | Founder & Elizabeth | Virtual COO (AI)
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.
