You bought the $20 AI plan for your employees. You think that means your company has integrated AI.
It probably has not.
The $20 plan may be enough for someone to try AI, ask occasional questions and become familiar with the technology. But if an employee is supposed to use AI throughout the working day and keeps hitting usage or capability limits, you have not given that person a serious business tool. You have given them a trial.
Then you look at a $200 plan and think it is expensive.
If sufficient AI access helps the right employee double, triple or quadruple what they can accomplish on suitable work, $200 is nothing.
It is closer to adding meaningful employee capacity for $200 a month than buying another piece of software.
If you cannot see how that additional capacity could happen, that is a different problem. Your company may not have found the right work, provided the right training or learned how to use AI properly yet. The problem is not automatically the price.
You bought access. You did not integrate AI.
AI integration is not measured by how many licences appear on your credit-card statement.
It shows up in the work.
Can an employee prepare a useful first draft in minutes instead of beginning with a blank page? Can the team respond to an inquiry while the opportunity is still active? Can someone change a workflow without waiting weeks for a software vendor? Can an employee test an idea visually before the next meeting?
If the work has not changed, AI has not been integrated.
The plan names differ between vendors, and an individual subscription is not the same as a managed business workspace. As of September 2026, Anthropic lists Claude Pro at US$20 per month and individual Max tiers at US$100 and US$200. OpenAI also lists individual options at those higher price points, while its managed Business plans use different seat prices and controls. The exact product may change. The distinction that matters is between occasional access and enough capacity to make AI part of real work. Sources: Claude plan documentation, ChatGPT release notes and the ChatGPT Business FAQ.
Not every employee needs the maximum plan. An employee who uses AI occasionally may be perfectly well served by an entry-level option. But the person expected to research, draft, analyse, build, revise and direct longer assignments cannot be evaluated fairly when the tool keeps stopping in the middle of the work.
That is like judging an employee's potential while giving them access to the office for only part of the day.
The $200 question is about output
The real question is: What can this employee accomplish with enough AI capacity that they cannot accomplish today?
If the answer is nothing, do not buy the larger plan yet. Find out why. The employee may need training. The workflow may be poorly chosen. The company information may not be ready for a governed knowledge assistant. The AI may be the wrong tool for the assignment.
But when an employee can direct AI across several useful activities, the economics change quickly. The employee is no longer using AI to polish one email. They are supervising additional productive capacity.
Research has already shown meaningful improvements on defined tasks. A field deployment involving 5,179 customer-support agents found about 14% more issues resolved per hour overall, with larger gains among less-experienced agents. In a product-development experiment, an individual working with AI produced solutions comparable in quality to a two-person team without AI on the studied assignment. Those findings do not prove that every employee becomes twice as productive across an entire job. They do show why it is reasonable to test AI as additional capacity rather than dismissing it as another software expense. Sources: Generative AI at Work and The Cybernetic Teammate.
At Salient AI, we have built and operated AI-assisted workflows for our own work and for clients. They keep evolving because each working use case reveals the next one.
What serious AI capacity can look like
AI can prepare replies for you and your staff, save them as drafts and let the responsible person review them before sending. In the right controlled situation, approved categories of messages could be sent automatically. The employee moves from composing every routine response to reviewing, correcting and handling the exceptions that require judgement. That is the operating idea behind a well-designed AI customer-support workflow.
Your website
AI can help create, update and reorganize the site as the business changes. Instead of every improvement waiting in a development queue, the organization can move much faster—with review, testing, version control and rollback still in place before consequential changes go live. Salient AI applies that model when building AI-enabled websites designed for ongoing change.
Social media
Imagine coordinating 250 posts in a week. A person could not sensibly create, format and organize that volume alone. An AI-assisted social-media pipeline can facilitate it while a human directs the strategy, checks the content and controls publication. The point is not to flood every channel. It is that work which was previously impossible for a small team becomes operationally possible.
The SaaS products you already use
How often has someone said, “I wish this tool could do one more thing”? In the past, the choices were usually to live with the friction, add another subscription or request a feature that might never arrive. AI-assisted development can make it practical to build or modify a focused internal tool quickly. The company can shape the workflow around the way its people actually work instead of permanently shaping its people around the software.
That does not make software engineering disappear. Security, maintenance, integrations, testing and ownership still matter. But the distance between “I wish it did this” and a working first version can become dramatically shorter. That is the type of friction we examine in Salient AI's work on administrative workflow automation.
A trade show or website inquiry
A person meets a promising contact, captures their information and records what they discussed. That contact can enter a sensible follow-up workflow immediately. AI can help prepare the relevant message, select appropriate supporting material and create the next action. A human can approve the communication where required instead of allowing the opportunity to disappear into a spreadsheet. A connected AI CRM workflow keeps the contact, conversation and next action together.
The same idea applies to an inquiry form. Someone fills it out on your website. Instead of receiving the same generic sequence as everyone else, the contact can enter a library of approved emails from which AI selects or adapts a suitable next message. The company can change that workflow as it learns, instead of waiting for a rigid nurture sequence to become obviously obsolete. Salient AI approaches responding to leads and building sales follow-up systems as one connected journey, not isolated messages.
Ideation
An employee does not need to arrive with a polished specification. They can describe a problem, explore alternatives and turn the idea into a visible workflow or interface. More people can participate because AI helps close the gap between having a thought and being able to show it to someone else.
These are not seven unrelated tricks. They are examples of an employee gaining the ability to communicate, build, test and operate with far more reach.
Human review is part of the system
None of this means AI should be allowed to act everywhere without oversight.
An email draft is different from an automatically sent email. A proposed website change is different from a production deployment. Suggested follow-up is different from contacting someone without valid consent.
Privacy, security, anti-spam law, contractual obligations and industry regulation can limit which information an AI system may access and what it may do. The consequence of an error matters too. A low-risk internal draft may need a quick review. A message involving health, employment, finance or another consequential decision may require a qualified person and a much stricter process.
The answer is not to avoid AI completely. It is to design the right boundary:
- What information may the system use?
- What may it prepare?
- What requires human approval?
- What, if anything, may it do automatically?
- Who remains accountable for the result?
Responsible AI is not separate from productive AI. Clear boundaries are what allow useful work to move quickly without pretending that speed removes responsibility.
The more people use AI, the more they can see
The most interesting use cases rarely arrive fully formed in the first meeting.
We may begin with email and discover that the larger delay happens after the reply. We may improve a website workflow and realize the same information should support sales follow-up. A client may try a system and notice an opportunity we did not see. We improve it together, and that improvement reveals another possibility.
The more our clients participate, the more they see what they had not seen before—and the more they help us see what we had not seen either.
That is human nature.
The more we play with something, the more we understand what it can do.
AI integration is therefore not a finished installation. It is a capability that develops through participation. That is why the first practical habits—such as learning to communicate with AI through a faster working conversation—matter more than a one-time demonstration.
I work with AI every day, and even I have to keep catching up. If I stopped using ChatGPT for one month, someone who continued using it throughout that month might become better than me. The technology changes that quickly, but so does the user's ability to recognize what is possible.
An employee confined to occasional use is not only producing less today. They are missing the practice that helps them identify tomorrow's use cases.
Your competitors are accumulating experience
This is the cost that does not appear beside the $200 subscription.
While you are deciding whether the larger plan is too expensive, another company may be learning how to answer customers faster, modify its systems sooner, test more ideas and remove more administrative friction.
Its advantage is not one magical automation. It is the accumulating ability of its people to see a problem and ask, “Could we do this differently now?” This is how the distance grows between owning AI licences and overcoming the institutional lag that keeps the operating model unchanged.
That makes the company faster, less expensive to operate and more agile. It can pursue opportunities your team does not have time to pursue. It can improve a workflow while your organization is still waiting for a vendor. It can learn from several experiments while you are still trying to justify the first one.
The gap compounds because output creates experience, and experience creates better ideas about where AI belongs next.
Give one employee enough capacity to prove it
Do not buy the most expensive plan for everyone tomorrow.
Choose one employee who already understands an important part of the business. Choose one recurring workflow with visible friction and a result you can inspect. Give that person sufficient AI access, practical training and permission to work differently.
Then measure something real:
- How much useful work was completed?
- How quickly did the customer or colleague receive a response?
- What could the employee do that was not practical before?
- Where did human judgement remain necessary?
- Which limitation disappeared when the employee had sufficient access?
If the larger plan creates no meaningful difference, learn from that evidence. The workflow, training, tool or plan may need to change.
But do not give an employee minimal access, prevent serious use and then conclude that AI does not create value.
The $20 plan can help someone begin. Integration starts when AI has enough capacity, knowledge, practice and responsibility to change what the employee can actually accomplish.
Written from Stephen Lau's experience and point of view, with AI-assisted research and editorial production. Reviewed by Salient AI before publication.