For thirty years, business software has worked the same way: a person opens a screen, finds the record, reads it, decides, and types the result into another screen. Every system you own, from the ERP to the CRM to the ticketing tool, assumes a human at the keyboard doing the connecting.
AI agents change that assumption, and by 2030 the change will be visible in how ordinary mid-size companies buy, build and use software. What follows is what we expect to change, what we expect to stay the same, and what to do about it now, based on the research and on the agents we are building today.
What an agent is
An agent is software that is given a goal, has access to tools and systems, and works out the steps itself: read the inbox, look up the customer, draft the reply, create the task. It differs from a chatbot, which answers, and from traditional automation, which follows a fixed script. We explained the distinction in chatbot versus AI agent and the business view in our guide to AI agents for business owners.
The agents we build for clients today are narrow and supervised: one that sorts inquiries into the CRM, one that gathers supplier quotes, one that chases invoices, one that drafts follow-up emails for approval. That narrowness is deliberate, and it is the shape of what works.
The hype, and the evidence against it
Expectations are running ahead of results. Gartner predicts that more than 40% of agentic AI projects will be canceled by the end of 2027, because of cost, unclear value and weak controls.
PwC's 2026 Global CEO Survey found that 56% of CEOs saw neither higher revenue nor lower costs from AI in the previous year. Our own research on AI adoption in mid-size businesses reached the same conclusion: the technology is further ahead than most companies' ability to use it.
None of that means agents will not matter by 2030. It means the winners will be the companies that treat agents as software projects with owners, controls and numbers, not as magic.
Five changes we expect by 2030
1. The screen stops being the main interface
Most routine work in business systems, data entry, lookups, status updates, retyping between tools, will be done by agents. People will spend their time on exceptions and decisions, and the software they use will be designed around approving, correcting and asking rather than filling forms. Expect the inbox, the chat window and the approval queue to become the main places work happens.
2. Systems get connected whether vendors like it or not
Agents need to read and write across the ERP, the CRM, the document store and the accounts. Companies whose systems do not expose data through APIs will find that their agents cannot do much. The integration work that firms have put off for a decade becomes urgent, which is why our advice to clients today starts with connecting systems rather than choosing a model.
3. Permissions and audit logs become the product
An agent that can create a purchase order needs limits: how much, from whom, with whose approval. Every action needs a record of what it did and why. By 2030 the differentiator between business software products will be how well they let a company control and audit what agents do inside them. We already build roles and audit logs into every SaaS product for enterprise buyers; agents make that non-negotiable for everyone.
4. Human-in-the-loop becomes the default design
The agents that survive the cancellation wave will be the ones that draft, propose and ask, with a named person approving anything that spends money, changes a record or goes to a customer. Our guide to human-in-the-loop AI explains the design. The ratio of approvals to autonomous actions will shift over time as trust builds, but the approval step will not disappear.
5. Custom becomes cheaper than off-the-shelf for the gaps
Packaged software will ship with agents for the common cases. The valuable agents will be the ones that know your specific process: how your firm prices a tender, how your plant decides what to make, how your clinic triages a message. Those may need custom integration or development, even as packaged tools improve. The decision framework in custom software versus off-the-shelf will tilt further toward custom for exactly these pieces.
What will not change
Data quality decides everything. An agent working from duplicate customers and stale stock figures makes confident mistakes. The firms that benefit will be the ones that fixed their data.
Accountability stays human. Regulators, customers and courts will hold a person responsible for what an agent did. The systems must make that person's approval visible.
Change is organizational. McKinsey's surveys keep finding that the firms getting value from AI redesign the work, not just add a tool. That was true of every previous wave of software and it will be true of agents.
What to do between now and then
Connect the systems that hold your customers, orders, stock and money, so an agent can read and write them. This is plumbing, and it is the biggest predictor of who adopts agents early.
Clean the data those systems hold. Our checklist for data readiness for AI is the place to start.
Build one supervised agent for a frequent, measurable process, and learn what it takes to run it. The experience transfers to the next five.
Write down who approves what. The rules are the hard part, and they take longer to agree than the code takes to write.
Related reading: Why AI agents stall at scale: a governance checklist for operations leaders
Frequently asked questions
Will AI agents replace our business software?
No. They will sit on top of it, doing the routine work inside it. The ERP, the CRM and the accounts package remain the systems of record; what changes is how much of the work in them is done by people.
Should we wait for our software vendors to add agents?
Vendors will add agents for the common cases. The ones that know your specific process will need to be built. Waiting for the vendor is reasonable for the generic work and a mistake for the work that makes your business different.
How do we stop an agent doing something expensive?
Limits and approvals. The agent has a budget, a scope and a list of actions it may take alone; everything else waits for a named person. Every action is logged. This is standard in the agents we build and it should be standard in anything you buy.
Is 2030 realistic for a mid-size company?
Firms already running supervised agents today may be able to expand them gradually by 2030. The firms that have not connected their systems will still be doing the plumbing. The difference is when the work starts, not the technology.
A sensible next step for AI agents
Pick one process you would hand to a reliable assistant tomorrow if you had one. That is your first agent. Our free 5-day AI readiness audit tells you whether your systems and data are ready for it, or book a 30-minute call and talk it through with a founder.
