From Law School to Agent Building: Reinventing a Career in the Age of AI
Eliott Ardisson
Founder & CEO - Basalt Studio
How professionals with legal, consulting, or marketing backgrounds are building AI agent careers — and what the transition actually looks like in practice.
Key Takeaways
- Domain expertise from law, consulting, marketing, or finance is a genuine asset when building AI agents — not a gap to apologize for
- Modern AI agent development does not require a computer science degree; the real skill is translating business problems into automatable workflows
- The professionals making the most successful transitions are those who start narrow: one industry, one pain point, one working prototype
- McKinsey and Gartner research both point to accelerating demand for people who can sit between technical AI systems and business operations
- The path is not linear or guaranteed, but the fundamentals — workflow thinking, stakeholder communication, problem framing — transfer directly from traditional professional careers
When Your Profession Starts Automating Itself
Here is the uncomfortable reality that most lawyers, consultants, and marketers are quietly sitting with: the tasks that consumed the bulk of junior and mid-level professional work — document review, client intake, research synthesis, lead qualification, expense reconciliation — are now within reach of automated systems. Not perfectly, not without oversight, but well enough to change the economics of how those tasks get staffed.
This does not mean professional careers are disappearing. It means the shape of valuable professional work is shifting. And that shift creates a specific kind of opportunity for the people who lived inside those workflows and understand them from the inside out.
Building AI agents is not a career reserved for engineers. At its core, it is a discipline of workflow design, business logic, and problem framing — skills that experienced professionals in law, finance, HR, and consulting have developed for years, even if they have never written a line of code.
What “Building an AI Agent” Actually Means
The phrase gets used loosely, so it is worth being precise.
An AI agent is a software system that can take a goal, break it into steps, use tools or data sources, and complete a task with minimal human intervention. In a business context, this might mean:
- An intake agent that receives a client inquiry, asks clarifying questions, extracts key information, and routes it to the right team member
- A document review agent that reads contracts, flags non-standard clauses, and produces a structured summary
- A lead qualification agent that scores inbound prospects against defined criteria and populates a CRM automatically
- A scheduling and follow-up agent that manages appointment reminders and no-show recovery for a service business
None of these require machine learning research or systems architecture. They require understanding the business problem well enough to define the inputs, the logic, and the expected output — and then knowing which tools to connect together to execute it.
The technical layer exists, but modern tooling has compressed it significantly. Platforms built around workflow automation, language model APIs, and pre-built connectors mean that a professional with genuine domain knowledge and a willingness to experiment can build functional agents without a computer science background.
Why Domain Expertise Is the Actual Advantage
This is the part that gets lost in most “career pivot to AI” narratives. The framing tends to be: go learn the technical tools, and you will be competitive. That is partly true. But what the market is genuinely short on is not people who know how to connect an API — it is people who understand the messy, exception-filled reality of how legal firms, recruitment agencies, accounting practices, or real estate brokerages actually operate.
A lawyer who has spent five years doing commercial contract review knows what a non-standard indemnity clause looks like, why it matters, and what the client needs to know about it. That context does not exist in a prompt template. It comes from professional experience, and it determines whether an AI-assisted review workflow is genuinely useful or merely fast.
A recruiter who has spent years screening candidates knows that job description requirements are often aspirational rather than literal, that certain industries have informal qualification signals that do not appear on CVs, and that timing in outreach matters as much as message quality. Building a candidate-matching agent without that knowledge produces something that technically works and practically misses the point.
The professionals who build the most useful agents are the ones who can define what good output looks like — because they have seen it, produced it, and been evaluated on it. That is not a technical skill. It is professional judgment, and it is the harder part to acquire.
What the Transition Actually Looks Like
There is no single path, but there are common patterns among professionals who make this shift successfully.
They start with a problem they have already lived. The first agent is usually personal: automating something tedious from their own previous workflow. A paralegal builds a clause-extraction tool for their own use. A marketing manager builds an agent that categorizes and routes inbound email inquiries. Starting here removes the pressure of a client relationship and lets them learn how the tools behave before the stakes are real.
They resist the urge to go broad too quickly. The instinct, especially among consultants, is to identify a large market opportunity and build a horizontal solution. This tends to produce something generic that nobody needs urgently enough to pay for. The agents that find early traction are narrow: one specific task, for one specific type of business, where the professional has direct credibility.
They learn by doing, not by accumulating certifications. The tools change fast enough that a course completed six months ago may already be covering deprecated functionality. The more durable learning happens in the process of building something that works — and debugging something that does not.
They treat early clients as collaborators, not evaluators. The first few implementations are as much about understanding what businesses actually need as they are about delivering output. Professionals who come in with humility about what they do not yet know, while bringing confidence in the domain they do know, tend to build better solutions and retain clients longer.
In our work helping founder-led SMBs deploy AI intake and workflow agents, the most common breakdown is not a technical failure — it is a scoping failure. Someone built an agent for the workflow as it was described, not the workflow as it is actually practiced. Domain insiders catch this earlier, because they already know where the exceptions live.
The Technical Floor You Actually Need
It is worth being honest about this, because the “no coding required” framing can mislead people into thinking the path is purely conceptual.
You do not need to build language models or write backend infrastructure. But you will need to become comfortable with:
- Prompt engineering: Understanding how to instruct a language model clearly, how to structure inputs, and how to handle edge cases in outputs
- Workflow logic: Thinking in terms of triggers, conditions, branches, and error handling — the same logic that underpins any automation tool
- API basics: Understanding what an API is, how to authenticate, and how to pass data between systems — even if you are using a no-code connector to do it
- Data handling: Knowing how structured and unstructured data differ, and how to extract, clean, and route information between steps
None of this requires formal training. Most of it can be learned by building, reading documentation, and asking language models to explain what you are looking at. The real prerequisite is a tolerance for ambiguity and a systematic approach to debugging when things do not work as expected.
Tools like n8n for workflow automation, the Claude API for language model capabilities, and TypeScript-based environments for more custom implementations are increasingly accessible to people who are willing to invest the time — even without a development background.
Common Points of Failure in Career Transitions
Overbuilding before validating. Spending three months building a polished agent for a use case nobody has confirmed they will pay for. The feedback loop needs to be short. Build the minimum version that tests the core assumption, then iterate.
Underpricing to compensate for imposter syndrome. Professionals coming from law, consulting, or finance often have strong market intuitions about value. When they pivot to AI work, they sometimes discount aggressively because they feel uncertain about the technical side. The domain expertise that makes their work genuinely useful is worth pricing accordingly.
Ignoring the change management problem. An agent that no one uses is not a solution — it is a demonstration. The hardest part of implementation is often not building the system but getting a team to adopt it. Professionals with consulting or operations backgrounds tend to be better at this than pure technical builders, but it still requires deliberate attention.
Chasing the latest model or platform. The tools landscape shifts constantly. Professionals who orient themselves around specific platforms become vulnerable to those platforms becoming obsolete. The more durable orientation is around the business problem: what does this client need to stop doing manually, and what does good output look like?
What Gartner and McKinsey Are Actually Saying
It is worth separating the signal from the hype here. The research firms are consistent on a few points.
Gartner has reported that agentic AI — systems that can take autonomous, multi-step actions — is among the top strategic technology trends for 2025 and beyond, with enterprise adoption accelerating significantly compared to earlier generations of automation tools.
McKinsey research suggests that generative AI could automate a meaningful share of work activities across knowledge-work sectors, with legal, financial services, and professional services among the most exposed categories. The nuance in their analysis — which often gets dropped in summarized versions — is that exposure to automation does not mean displacement. It means restructuring: the nature of the work changes, and the professionals who adapt fastest are positioned to capture more value, not less.
Deloitte and Accenture have both published research pointing to a growing gap between the demand for AI implementation capability and the supply of people who can translate between business requirements and technical execution. That gap is the opportunity. It is not infinite, and it will narrow as the field matures — which is the case for most new disciplines.
A Realistic View of the Timeline
Building a functional first agent: two to six weeks, depending on complexity and how much of your time you can dedicate to it.
Developing enough range to take on paid client work: three to nine months of consistent practice and iteration.
Building a reputation and a pipeline in a specific niche: one to two years, for most people.
The six-figure consulting practice in twelve months narrative exists, but it is the exception, not the baseline. The more useful mental model is: this is a skill acquisition process with real market demand at the end of it, and the professionals who approach it with patience and specificity tend to build more durable practices than those chasing the fastest possible revenue.
Choosing Where to Start
If you are evaluating this transition, the most useful first question is not “what tools should I learn?” It is: “What business problem do I understand well enough to build a solution for, and who would actually pay to have it solved?”
The answer to that question comes from your professional history, not from the AI tool landscape. Once you have a specific problem and a specific type of client in mind, the tools question becomes much easier to answer — because you can evaluate them against a concrete use case rather than in the abstract.
There are many ways to structure the learning process: building independently, working alongside an implementation shop on a client engagement, or starting with a consulting role that involves AI scoping before moving into building. What matters less than the path is the specificity of the focus. Narrow beats broad, especially in the early stages.
Where to Go From Here
The professionals who will build the most durable careers in AI implementation are not the ones who learned to code fastest — they are the ones who understood a business domain deeply enough to know what actually needed to be automated, and why.
If you are a founder or operator thinking about where AI implementation fits into your business, or a professional evaluating whether this kind of work makes sense for your next chapter, a focused conversation is often the fastest way to cut through the noise.
You can book an AI strategy call with the Basalt Studio team at https://cal.com/eliott-ardisson-kzq7zs/ai-strategy-call — no pitch deck, just a direct conversation about what is actually feasible for your situation.