“Don’t Be Afraid of Technology”: One Builder’s Path to Purpose-Driven AI
Eliott Ardisson
Founder & CEO - Basalt Studio
What purpose-driven AI actually means for founder-led SMBs: why starting with the problem — not the tool — leads to automation that sticks.
Key Takeaways
- Purpose-driven AI starts with a specific problem to solve, not a technology to deploy. The order matters.
- Small, targeted implementations in one workflow consistently outperform broad, multi-system overhauls for SMBs with lean teams.
- The most durable AI adoption happens when automation removes work people find draining, not work that gives them professional identity.
- Non-technical founders can implement meaningful AI agents today. The barrier is process clarity, not programming ability.
- Human judgment and contextual decision-making remain the core of good work. AI earns its place by protecting time for both.
The Question Most Founders Get Backwards
When a founder-led business decides to explore AI, the first question is usually some version of: “What tools should we be using?” It’s the wrong starting point.
The right question is: “Where does our team spend time on work that doesn’t require human judgment?” That reframe sounds simple. In practice, it changes everything about how an implementation gets scoped, built, and adopted.
Purpose-driven AI is a term for this approach. It means the problem definition comes before the technology selection. It means the design of an agent is constrained by what would actually be helpful for a specific team, not by what is technically impressive or what a vendor is selling. And it means the measure of success is whether the people using it work better, not whether a dashboard shows a usage number climbing.
This post is about what that looks like in practice for founder-led SMBs — the kinds of businesses running 10 to 100 people in professional services, real estate, recruitment, legal, HVAC, and similar fields where most revenue depends on relationships and expert judgment.
What “Purpose-Driven” Actually Means in Practice
The phrase can sound abstract, so it helps to anchor it in contrast.
Traditional automation is transactional: if a form is submitted, send a confirmation email. If a row is added to a spreadsheet, notify a Slack channel. These are useful, but they are brittle. They break when the input is slightly off-format. They don’t handle exceptions. And they don’t get smarter over time.
Modern AI agents are different. They can read context, handle variation in inputs, make conditional decisions, and route tasks appropriately. But the capability is not the point. The point is whether that capability is aimed at something worth solving.
A purpose-driven approach asks three questions before any technical work begins:
What specific task currently consumes disproportionate time? Not “admin in general,” but something concrete. Qualifying inbound leads. Drafting first-pass responses to standard client queries. Extracting line items from supplier invoices and logging them to a tracking sheet.
What would a good outcome look like for the person doing that task today? This matters because good AI design serves the human in the loop, not just the business owner looking at efficiency metrics. If a recruiter spends three hours a day screening CVs, a good outcome might be arriving at a shortlist of ten candidates with brief notes, rather than having the AI rank everyone with a score they don’t trust.
What does the AI need to know to do this reasonably well? This is where most implementations fail or succeed early. If the context required to do the job well is too implicit, too tacit, or too variable, the AI will produce outputs that need so much correction they create more work than they save.
When these three questions have clear answers, you have the foundation for something worth building.
Why Personal and Operational Experience Beats Technical Expertise
Some of the most useful AI agents running inside small businesses today were designed by people who aren’t developers. They were designed by people who understood a workflow intimately because they had lived it.
This matters because AI implementation is fundamentally a design problem, not an engineering problem. The hardest part is not connecting an API or writing a prompt. The hardest part is knowing what the output should look like, what edge cases need to be handled, and what the system should do when it is uncertain.
A legal firm’s office manager who has been drafting client intake summaries for eight years knows things about that task that no outside developer will discover in a two-hour discovery call. She knows that clients from certain referral sources tend to have more complex situations. She knows that some questions on the intake form get answered inconsistently and need a follow-up. She knows which pieces of information the solicitor actually reads first.
If that knowledge gets incorporated into how an intake agent is designed, the agent produces something genuinely useful. If it doesn’t, the agent produces something technically functional but practically ignored.
This is why the best AI implementations in small businesses tend to emerge from operators who are close to the work, not from top-down technology mandates. And it is why the “don’t be afraid of technology” framing matters. Modern tools are accessible enough that people with deep process knowledge can guide and shape implementations, even without writing code themselves.
Where AI Fits Well in Founder-Led SMBs
Not every workflow benefits from automation. The ones that do tend to share a few characteristics: they are repetitive, they follow recognizable patterns, they are time-consuming relative to the decision complexity involved, and they don’t require the kind of relational trust that clients pay a premium for.
Here are some practical examples across the industries where this tends to work well.
Recruitment and HR: Initial CV screening, interview scheduling, follow-up communications with candidates at different stages, and generating structured summaries from unstructured interview notes. These tasks can consume hours per day in a busy agency. An AI agent handling first-pass triage allows consultants to spend their time on conversations, not administration.
Legal and professional services: Document review for standard clauses, drafting first versions of routine correspondence, extracting key dates and obligations from contracts for a matter management sheet, and responding to standard client status enquiries. McKinsey research on professional services has consistently pointed to document-intensive work as one of the highest-leverage areas for AI-assisted productivity.
Real estate: Lead qualification from inbound enquiries, generating property description drafts from structured listing data, following up with buyers at defined stages of the pipeline, and compiling comparable sales data for agent review.
HVAC and trades: Job scheduling coordination, follow-up for service reminders, generating quotes from standard job types, and processing supplier invoices. These businesses often have significant administrative overhead relative to team size, and the work is highly pattern-based.
Accounting and financial services: Data extraction from client documents, reconciliation checking, drafting client communications for standard reporting periods, and flagging anomalies in financial data for human review.
In each case, the AI handles the pattern-recognition and production work. The human handles the judgment, the relationship, and the exceptions.
The Common Breakdown Points
In our work helping founder-led businesses deploy AI agents across these kinds of workflows, the most common breakdown is not technical. It is definitional.
Teams often enter an implementation without a shared understanding of what “done” looks like. What does a good AI-generated CV summary include? How should the agent handle an edge case where a candidate has an unusual background? What tone should client-facing messages use? When should the agent escalate to a human rather than proceed?
These questions need answers before development starts, not during testing. When they get deferred, the result is an agent that works technically but requires so much manual correction that adoption collapses.
The second common breakdown is scope creep. A founder will start with a clear problem — qualifying inbound leads — and during the build process, add requirements: “Can it also send the calendar invite? Can it also pull their LinkedIn profile? Can it also update the CRM?” Each addition is reasonable in isolation. Collectively, they extend timelines, introduce integration complexity, and delay the moment when anyone sees results.
The discipline of starting with one well-defined workflow, running it to completion, and measuring it before expanding is not a limitation. It is what makes the difference between an AI implementation that gets used and one that gets abandoned.
How to Approach the Mindset Shift
The “don’t be afraid of technology” framing resonates because the fear is real. Founders worry about disrupting processes that work, even imperfectly. They worry about team resistance. They worry about building something that breaks at a critical moment. They worry about becoming dependent on a system they don’t fully understand.
These are reasonable concerns, not obstacles to dismiss. The answer to each of them is the same: start smaller than feels meaningful.
Pick a workflow that, if the AI gets it wrong, causes inconvenience rather than damage. A first-draft email that a human reviews before sending. A data extraction task where the output gets checked against the source document. A lead qualification filter where the sales rep sees the AI’s reasoning before acting on it.
This bounded scope serves two purposes. It limits exposure while you learn what the system does well. And it builds the kind of experiential trust that no amount of vendor reassurance can provide. When a team member has watched an agent handle 200 inbound enquiries and seen that it handles them reliably, their relationship with that tool is fundamentally different from someone who was told it works.
Gartner has noted in its research on enterprise AI adoption that teams which pilot AI in low-stakes environments before expanding to higher-stakes workflows show significantly better long-term adoption rates. The same principle applies at the SMB scale.
What Good Implementation Actually Looks Like
A well-run AI implementation for a founder-led SMB typically moves through a few recognizable phases.
First, there is a workflow audit. This is not a technology conversation. It is a process conversation: what happens, in what order, who touches it, where does it slow down, what does a good output look like, what does a bad one look like. This phase produces the specification that the AI agent is built against.
Second, a single agent gets built and tested against real data. Not synthetic examples. Real emails, real documents, real enquiries from the last quarter. This is where assumptions get challenged and the specification gets refined.
Third, the team that will use the agent is trained — not on the technical internals, but on what the agent does, what it doesn’t do, how to review its outputs, and how to flag problems. Change management at this scale is not complicated, but skipping it reliably produces poor adoption.
Fourth, the agent runs with oversight. Someone monitors outputs for the first few weeks, catches systematic errors, and adjusts the configuration. This is normal. No agent is correctly calibrated on day one.
Fifth, once the first workflow is stable and measurably useful, the scope expands to the next priority.
This is not a dramatic transformation. It is a methodical addition of capacity to an existing team. That is exactly what it should feel like.
The Principle That Holds Everything Together
The most important thing about purpose-driven AI is also the simplest: technology is not the goal. Useful work, delivered by capable people with more time and less friction, is the goal.
AI earns its place in a business by making that outcome more achievable. When it does, teams adopt it and build on it. When it doesn’t, it becomes shelfware. The technology is neutral. The design decisions are not.
For founders thinking about where to start, the practical question is: where does your team spend the most time on work that follows predictable patterns? Start there. Define the output you want. Build something small. Watch it work. Then go from there.
If you want to map out which workflows in your business are the right candidates for AI implementation, Basalt Studio offers a focused AI strategy call where we work through your current processes and identify the highest-leverage starting point. No pitch deck, no commitment required.