I recently spoke with a solo-practitioner attorney who told me she had “implemented Artificial Intelligence (AI) everywhere.” Given how aggressively the legal industry cheerleaders are pushing AI-led transformation, I expected something sophisticated, perhaps an agentic system handling contract negotiations or a custom-trained legal model. Instead, she described something far simpler:
- Gemini to draft emails and summarize contracts
- NotebookLM to organize expert testimonies
- ChatGPT to help write blog posts
That was it. No agents. No orchestration layers. No “AI platforms.”
Reach out to discuss this topic in depth.
And yet, she told me it was saving her enormous amounts of time, which if you are a solo practitioner, is the single most valuable commodity. When I asked why she started there, the answer was blunt: “Because it was easy to implement”, and now she was looking into a more complex Contract Lifecycle solution.
This answer exposes a problem with how most enterprises are approaching AI.
The enterprise AI delusion
Across industries, Chief Information Officers (CIOs) are being told the same thing: “we need agentic AI, and we need it now.” The implication is clear:
- Incremental use cases are too small
- AI productivity tools are now table stakes
- Real value comes from full-scale AI-enabled transformation through Agentic AI.
This is wrong. In fact, most enterprises pursuing top-down agentic strategies, at the expense of not driving bottom-up workers’ productivity are making a fundamental mistake – they are trying to run before they can walk.
Agentic AI sits in the most complex, least proven quadrant of the implementation complexity matrix (see exhibit 1), which requires:
- Re-architected business processes
- New governance models
- Observability into probabilistic systems
- Tolerance for failure that most enterprises are not equipped to manage
Yet organizations are jumping there first, inevitably encountering endless pilots and very little scaled value.
Exhibit 1:
There’s a growing body of quiet evidence that should concern every CIO:
- AI pilots stall because they are bolted onto workflows never designed for it
- Agent-based systems break in unpredictable ways and are hard to debug
- Costs spiral due to unpredictability of token usage
- Governance teams cannot keep up with decentralized experimentation
Meanwhile, something else is happening, largely ignored by leadership: employees have already adopted AI and use productivity tools outside official channels.
In other words, while enterprises are investing millions in top-down AI transformation, the bottom-up AI adoption is already delivering measurable productivity gains directly to the employees.
A more effective path: Learn before you leap
To arrive to business-wide agentic transformation, enterprises should deploy both strategies:
- Top-down – running targeted Agentic AI experiments, and
- Bottom-up – allowing evolution and a staged capability build-up through Learning Paths (see exhibit 2).
The IT Learning Path would start with reinvention of how Information Technology (IT) operates, before entering the business domain. For example:
- Begin with AI-assisted coding to increase developer productivity
- Automate testing and debugging workflows
- Start experimenting with self-healing infrastructure
This is where teams learn:
- how AI behaves in real systems
- how to monitor and govern it
- how to manage failure
Most importantly, it is where organizations confront the realities of “agentic” behavior without putting core business processes at risk.
While IT builds capability, the business may start harnessing the employee “shadow adoption” by starting up the Business Learning Path:
- Provide secure, enterprise-grade “thinking assistants” to help with basic day-to-day tasks
- Allow AI-driven productivity tools to scale and observe how work patterns change
- Evolve from AI assistants helping with personal productivity to AI agents running business workflows
When employees adopt AI they break the existing cultural models (e.g., the rise of “fluid calendars” based on AI assistants). These behavioral changes become precursors to real transformation and ignoring them is a mistake. By the time an organization reaches agentic AI, it should already have:
- Technical capability (from IT Learning Path)
- Behavioral readiness (from Business Learning Path)
- Governance frameworks (from managing real usage in both cases)
Exhibit 2:
For CIOs, the imperative is not to win the race to the most advanced AI deployment, but to win the race to sustainable value. Methodically building the right capabilities to support sophisticated AI systems and compounding gains is a better way to pull ahead than leaping into full-scale agentic transformations. While the early bird often gets the worm, it’s the second mouse that gets the cheese.
The final thoughts
The solo attorney did not “implement AI everywhere.” She did something far more important: she prioritized learning and built confidence and capability first. It was not a compromise strategy; it was the one most enterprises are still missing.
If you enjoyed this blog, check out, Agentic AI: True Autonomy or Task-based Hyperautomation? – Everest Group Research Portal, which delves deeper into another topic relating to artificial intelligence and Agentic AI.
If you’d like to continue this discussion, please contact Ross Tisnovsky ([email protected]).