A language school with two admissions advisers spends most mornings the same way. Someone opens the inbox, answers the same questions about course dates and fees, copies details into a spreadsheet and promises to call back. Now a stream of vendors wants to sell the school an AI agent to do that work.
Does that make the school's software pointless? Not really. What changes is who presses the buttons. With most software today, a person logs in and does each step. With an AI agent, the person sets the goal and reviews the exceptions. The records and rules underneath matter more than before, not less.
Why everyone is talking about the end of SaaS
The word SaaSpocalypse is everywhere this year, and analysts have now put numbers on it. In July 2026, Gartner estimated that up to $234 billion of enterprise application spending is exposed to what it calls agentic arbitrage by 2030, roughly 20% of enterprise application SaaS spend. In plain terms, an agent completes tasks across several systems, so people need each screen less.
The same release is calmer than the headlines. Gartner called the change a metamorphosis rather than an apocalypse, and said SaaS will take a different form instead of disappearing. What changes is how you use software and how it's priced.
What actually changes when an agent does the work
Here is the school's enquiry desk as a hypothetical example.

| Task | Today | With an AI agent | Still needs a person |
|---|---|---|---|
| Course dates and fees | Copied from a document | Answered from approved information | When prices change |
| Student level and schedule | Asked over several messages | Collected in one exchange | Unclear answers |
| Placement test booking | Checked in a calendar | Open slots offered | Disputes |
| Discounts and refunds | Adviser decides | Never decided by the agent | Always |
| Complaints | Adviser replies | Acknowledged and handed over | Always |
Look at the last column. The agent takes the repetitive steps and the judgment calls stay with people, so the inbox and customer record matter more. When the agent hands over, the adviser needs the full history, the details collected and a clear owner. Otherwise the student repeats everything and the time saved disappears.
Gartner made the same point, noting that better results from AI depend on keeping customer context over time. In practice that means one place where conversations and ownership live, such as a shared inbox with the customer journey attached to each contact.
Be careful with anything called an agent
Plenty of products have simply been renamed. In June 2025, Gartner warned about agent washing, the rebranding of chatbots and assistants that lack real agentic capability. It estimated that only about 130 of the thousands of agentic AI vendors are real, and predicted that over 40% of agentic AI projects will be canceled by the end of 2027 because of rising costs, unclear value or weak risk controls.
- What can it do aloneList the actions it takes without a person approving them.
- What happens when it is unsureIt should hand over to a named person instead of guessing.
- Can you see every actionEach reply, booking and change should appear in a log.
- What does a failure costCheck whether failed or repeated attempts are still billed.
Gartner suggests AI agents where a decision is needed, plain automation for routine workflows and assistants for simple lookups. A reminder sent the day before a placement test is automation, and that's fine.
The checklist above covers what to ask. Our scorecard for evaluating customer service AI turns the answers into a test on your own conversations.
How pricing is starting to follow the work
Seat pricing made sense when software was a screen. When an agent does the work, many vendors charge for the work, per conversation, resolution or action. That can be fair, but bills get harder to predict.
With DripTell we kept monthly plans with a message allowance. AI replies use the allowance set for the workspace, and AI Calls run on prepaid credits with no automatic paid top-ups. DripTell adds no automatic overage charges, though WhatsApp message fees are billed separately.
Where to start this quarter
Pick one repetitive task, such as answering course dates. Write the approved answers down. Decide exactly when the agent hands over and who receives the conversation. Then measure the result for a month, as our guide to building an outcome ledger describes. If it works, add the next task.
Frequently Asked Questions
Is SaaS dead because of AI agents
No. Gartner expects SaaS to change form, with more work done by agents and less reliance on seat pricing. You still need records, permissions and people reviewing exceptions.
What is agentic AI as a service
Software that completes multi-step tasks toward a goal you set, instead of waiting for a person to click through each step. A person still reviews unclear cases.
How can I tell a real AI agent from a renamed chatbot
Ask what it decides alone, what it does when unsure and whether every action is logged. A chatbot that answers fixed questions is useful, but it isn't an agent.
Want help applying this guide?Ask the DripTell team
From reading to action
Want help applying this to your business?
Share your details. We will read the article you are on and start with a useful answer for your team.
DripTell Editorial
Practical guidance reviewed by the DripTell product and customer workflow team.
See how DripTell checks product claims, uses primary sources and handles corrections.
Editorial and source policy



