People like to say they have a very particular set of skills. The awkward part, from the conversations I keep having, is that those skills now look like about six job descriptions.
I have been talking with quite a few marketing agencies recently. The same observation comes up. Some of their specialists would rather stay inside the work they already trust, even when AI makes it much easier to cross into the next discipline. They know their field. They trust the years they have already put in. Doing the job that used to belong to somebody else is not automatically interesting to them.
Smaller teams, and some larger ones, are moving the other way. They want people who understand several disciplines well enough to connect them, and who can get more of the work done with AI. I think that makes sense. It also has a limit, and the limit is the part worth hiring against.
For a normal business website, a skilled generalist who understands how the pieces fit can now cover much more of the work. When one part has to be exceptional, an experienced specialist is still the better investment. The strongest hire is often the generalist who can tell you which is which.
A business website is several jobs with one outcome
If you are hiring someone to build the site, you are rarely hiring for a single craft. You need a design people trust, a build that holds up, a content structure that matches how you actually sell, SEO so the right people can find it, performance so the site does not feel broken, analytics so you can see what happened, and sometimes a connection to the CRM or marketing tools you already pay for.
Traditionally that meant several specialists, an agency coordinating them, or one freelancer trying to cover all of it and quietly dropping the parts they were weakest at. Coordination has a real cost. I broke down what that does to a budget in what a small business website actually costs. After launch, marketing the site is another system sitting on top of the same pages.
AI did not delete any of those jobs. It changed how much of the ordinary version of each job one prepared person can do in the same week.
What AI changes for a generalist
A skilled generalist who already understands design, development, SEO, analytics and marketing can move between those fields faster than they could a few years ago. AI helps with the research, the implementation, the troubleshooting, and with getting enough working knowledge to attempt a problem that used to require another person in the room.
There is a proper experiment behind the hopeful version of that claim, and it is more specific than the usual talk.
In 2024, the BCG Henderson Institute ran a controlled study with consultants doing work outside their normal job. 480 consultants completed the experiment, and 44 BCG data scientists set the benchmark without using the tool. The tasks were short and concrete: write Python to merge and clean two data sets, build a predictive model, and check statistical output that ChatGPT had produced. Each task was limited to 90 minutes. Half the consultants used Enterprise ChatGPT. Half were told not to.
On the coding task, the consultants using AI scored 86% of the data-scientist benchmark. That was 49 percentage points above the consultants working without AI, and they finished about 10% faster than the data scientists. Consultants who had never written code still reached 84% of that benchmark once they had the tool. People without it often barely got past opening the files. They landed at 29% of the benchmark.
That is a real result. It is also a narrow one. These consultants already knew what a correct data-cleaning result should look like, even when they could not write the Python. BCG is explicit that doing the task with AI did not mean they had learned the field. On a follow-up test, people who had just done the coding task scored about the same as people who had not done it. Ninety minutes with the tool did not install a new skill. BCG’s own description was an exoskeleton: useful while you are wearing it.
The predictive-modeling task is the part I would want a buyer to remember. There, neither the consultants nor the model were especially strong, and people with AI were more likely to be led astray. They were 15 percentage points more likely to select and apply a machine-learning method appropriately, and they were still the group least likely to match a data scientist. If you cannot tell that the answer is aimed at the wrong goal, speed gets you to the wrong place sooner.
So a prepared generalist can walk into unfamiliar technical work and produce something useful. The study does not say that person has become the specialist. BCG says the same thing in plainer language: they are not suggesting a nontechnical generalist can immediately become a data scientist.
You can finish the task and still miss the field
Every serious discipline has become crowded. A generalist today cannot hold expert-level knowledge in design, engineering, SEO, analytics and marketing at the same time. There is too much of it, and it keeps moving.
The earlier Harvard and BCG field experiment is the cleanest description I have seen of that limit. Published in Organization Science, it followed 758 BCG consultants. The public working paper is the version you can actually read. On 18 realistic tasks inside what GPT-4 could reliably do, people using the model completed 12.2% more tasks, finished 25.1% faster, and produced significantly better work. On one managerial task chosen to sit outside that range, the people using AI were 19% less likely to produce a correct solution than the people working without it.
Same kind of person. Same class of tool. The result flipped with the task.
I see the same split on websites. AI will happily build a page, a schema block, a form-to-CRM flow, or a rewrite of your service copy. Some of that output is fine. Some of it is confident and wrong in a way that only shows up after a customer, a search engine, or your sales process hits it. I wrote separately about why buying the tool does not buy the judgment.
Can a skilled generalist enter unfamiliar territory, find what matters, learn quickly, and ship something useful? Yes. I do that. Can they match someone who has spent fifteen years on the ugly problems in that one discipline? Sometimes, if the specialist stopped learning years ago. Against someone who is still sharp, still in the details, and also using AI well, I would not bet on the generalist for that specific problem.
A job title is a weak stand-in for expertise
A degree, a job title, or ten years in a role does not automatically make someone outstanding. An intellectually curious generalist who spends a few months studying a subject seriously can outperform a specialist who stopped developing. I would rather hire the person who is still learning.
Put that same generalist against a genuinely strong specialist, someone who keeps improving and knows how the work fails in real conditions, and the comparison changes. If that specialist uses AI well too, you get depth and speed together. That is a difficult combination to imitate with a prompt and a weekend of reading.
I also would not treat “specialist” as a synonym for “reluctant to use AI.” Some of the reluctance I hear is real. People protect the work they are known for, and they are not excited about being asked to become a second department. A 2022 survey of 164 medical students and 42 physicians found that threats to expert status and to professional autonomy contributed to resistance toward AI. That was medicine, with a vignette of a medical AI system, so I am not going to stretch it into a law of marketing agencies. It does line up with the conversations. The friction is often about identity and control, with the software sitting underneath that.
The closer evidence for marketers points somewhere more ordinary. In the 2025 State of Marketing AI Report, 62% of respondents named lack of education and training as a barrier to adoption. It was the top barrier for the fifth year in a row, ahead of lack of awareness. Fear or mistrust was on the list too, at 35%. A lot of people are stuck because nobody taught them. That is a different problem from a specialist who has looked at the tool and decided their existing judgment is still the valuable part.
Experience is the slow part
You can acquire information quickly. You can learn a method, understand a system, and generate a solution that is technically correct. Years of unusual problems, mistakes, pattern recognition and judgment are much harder to reproduce. AI is good at the first list. It is weak at the second, because the second is mostly scar tissue from having been wrong in specific situations.
I do not think specialization is dying. Look at how technology usually goes, from farming and construction through engines, energy and computing. The interface gets simpler. The system underneath gets more complicated, and more tied to everything else around it. AI can make the interface easier to operate. Understanding the system is still a different job.
For a small or medium-sized business, a skilled generalist with AI is an increasingly sensible way to get a website built. Less coordination, often a lower cost, and one person who can see whether the design, the content and the measurement are even talking about the same offer.
If the project needs exceptional performance in one place, I would still pay for the specialist. Advanced security. A messy integration. SEO in a market where a lot of competent people are already competing. Engineering that has to survive real traffic and real edge cases. Connecting a website to a CRM is a good example of work that looks like “just send the form” until the data, the ownership and the follow-up are wrong.
Sometimes the best option is the generalist who can do the bulk of the site and can tell you, specifically, when the specialist is worth the extra money and the extra coordination.
Who I would hire for this website
None of these is universally better. I compared freelancer, small team, agency and DIY routes in more detail in who should actually build the website. The split below is the one I use when the question is breadth versus depth.
| Multidisciplinary freelancer | Small team | Specialist or specialist agency |
|---|---|---|
| Fits a normal business site: design, build, content structure, basic SEO, performance, analytics, a straightforward integration. | Fits when design, build and marketing all need to move at once and one person would be the bottleneck. | Fits when one part is the actual risk: security, a complex integration, competitive search, or custom engineering. |
| You get one person who can make the pieces agree, with less handoff. | You get breadth split across a few people, which only helps if they actually talk to each other. | You pay for someone who has already been wrong in that specific kind of work and adjusted. |
| Weak when they cannot see their own limits and keep the hard part anyway. | Weak when you are mostly paying for coordination and still getting generic pages. | Weak when the site is ordinary and you are paying specialist rates for straightforward work. |
What to look for in the person
Work you can inspect
Ask what they actually built, and what they left to someone else. A live site beats a skills list.
A decision you can follow
They should be able to explain why they chose a layout, a platform, or a scope cut, in plain language.
A stated limit
The credible generalist can name the problem where they would bring a specialist in.
Pieces that agree
Design, content, SEO and measurement should describe the same offer. If they do not, breadth was only a list.
Hiring mistakes I keep seeing
The longest skills list
Design, development, SEO, ads, branding and automation on one profile is cheap to write. It does not tell you what shipped.
Ten years as current skill
Time in a role can mean deep judgment. It can also mean the person stopped updating the judgment years ago.
AI output treated as finished
A page that looks complete can still be aimed at the wrong customer, the wrong query, or the wrong next step.
The wrong shape of team
A room of specialists for a straightforward service site, or one generalist for a security-critical system, both start from the wrong risk.
What I would pay attention to
If I were hiring someone to build a business website, I would spend less time counting skills and more time on what they have shipped, how they explain a decision, and whether they understand where they should stop. The practical checks for that conversation are in how to hire a web developer when you are not technical.
A very particular set of skills is useful. Knowing what to do with them, and when to hand a piece to someone stronger at it, matters more.
If you already know what the site needs to do, send me the requirements and I will tell you whether I would keep the project with one person or bring a specialist in for a specific part.
Frequently asked questions
Can one person build a business website with AI?
A skilled generalist can cover design, development, content structure, basic SEO, performance and a straightforward integration on a normal business site. The limit shows up when one part needs years of specialist judgment, or when nobody is checking whether the AI output actually fits the business.
When should I hire a specialist instead?
Hire a specialist when the project depends on exceptional work in one area: advanced security, a complex integration, competitive SEO, or engineering that has to hold up under real use. A straightforward service website rarely needs a separate person for every discipline.
Does the research say generalists are replacing specialists?
No. The 2024 BCG experiment showed consultants using AI could perform some data-science tasks outside their training, especially coding, and still fall short of data scientists where judgment mattered. An earlier Harvard and BCG study found AI made people less likely to be correct on a task outside the tool’s reliable range.
How do I tell a real generalist from a long skills list?
Ask what they have shipped, how they decided something, and where they would bring someone else in. A list that includes design, development, SEO, ads and branding is easy to publish. A clear example of those pieces working together is harder to fake.
Are specialists refusing to use AI?
Some are, and some are not. A medical study found that threats to expertise and autonomy can feed resistance. Among marketers, the 2025 State of Marketing AI Report found lack of training was a more common cited barrier than fear. Judge the person in front of you. The job title is a weak signal.
