Automation is about to change shape
For two decades, business process automation has automated tasks; artificial general intelligence (AGI) would automate judgment. That is a bigger shift than any tooling upgrade of the RPA era, and it will change how automation is designed, delivered and governed.
AGI, meaning AI that can learn and reason across domains at roughly human level, does not exist today. But today’s AI agents already point in its direction, and automation leaders who wait for a formal “AGI moment” will be late. This post covers where automation stands, what AGI would change, and how to prepare now.
Where automation stands today
Rule-based automation works well on stable, structured processes and struggles everywhere else. Bots follow scripts: click here, read this column, copy that value. That delivers real savings on high-volume work such as invoice posting or report generation, but it carries known costs:
- Brittleness: a moved button, a renamed column or a new file layout can break a bot.
- Maintenance load: a large share of a mature team’s effort goes into fixing and updating bots rather than building new ones.
- Exceptions: anything outside the script lands back with a person.
- Slow onboarding: each new process needs detailed documentation and custom logic before it runs.
The more variable the inputs, the weaker the business case for a scripted bot. That is the gap AGI aims to close.
From bots to agents to AGI
Automation has moved through generations, and each one hands more of the thinking to the machine.
| Generation | How it works | Strong at | Main limit |
|---|---|---|---|
| Scripted RPA | Follows fixed rules and UI steps | Stable, structured, high-volume tasks | Breaks when screens or inputs change |
| AI-assisted automation | Adds OCR, document understanding and ML models to bots | Semi-structured documents, classification | Still needs a human-designed flow and narrow models |
| Agentic AI | Language models plan steps, use tools and handle exceptions | Variable inputs, multi-step work under supervision | Uneven reliability, needs guardrails |
| AGI (hypothetical) | General reasoning and learning across domains | Owning an outcome end to end | Does not exist today; timing is uncertain |
The real difference at the AGI step is what you hand over. Instead of automating a documented process, you hand over the goal and let the system work out the process.
Five ways AGI would reshape process automation
- Design shifts from documenting to describing. Instead of writing a step-by-step process document, teams state the outcome and the business rules, and the system works out the steps.
- Exceptions stop being the bottleneck. Today they go back to people. A general system could reason through an unfamiliar case, check policy and decide, escalating only genuine judgment calls.
- Automation spans whole processes, not tasks. Order-to-cash or procure-to-pay, which cross many systems and teams, become one automatable unit instead of a chain of separate bots.
- Maintenance largely fades. A system that understands intent can adapt when a screen, file layout or sign convention changes, rather than waiting for a developer to patch it.
- Decisions get automated, not just actions. Approvals, reconciliation matching and credit checks move from a human in the loop to a human on the loop.
What it means for automation teams
Roles shift from building bots to directing and assuring automated work.
- RPA developers: less scripting of selectors and flows, more designing agent behavior, testing, and integrating with enterprise systems such as SAP.
- Business analysts: process knowledge becomes the key input. Defining rules, policies and success criteria matters more than documenting every click.
- Project managers: delivery moves from build-heavy timelines to continuous rollout, with governance, risk and change management at the center.
The skill that keeps its value across all three roles is a deep understanding of the process itself.
Risks and governance
More autonomy raises the stakes, so four areas need an owner before any rollout:
- Accuracy and auditability: every automated decision must be explainable and traceable, especially in finance, payroll and banking.
- Accountability: a named person must own the outcomes the system produces.
- Security and access: an agent that acts across systems needs least-privilege access, careful credential handling and full activity logs.
- Compliance: regulated processes need approval gates, segregation of duties and evidence ready for audits.
Autonomy should be earned in steps. Start with human approval on every action, and relax it only where measured error rates justify it.
How to prepare now
- Map your processes. Record volumes, exception rates, input variability and owners, so you know where scripted bots struggle.
- Run a focused pilot. Pick a variable, document-heavy process and test an AI agent with human review of every output.
- Build governance early. Define approval rules, logging, access control and risk tiers before autonomy increases.
- Invest in integration. Governed APIs and clean access to systems like SAP and Excel beat screen scraping for any future system.
- Reskill the team. Move developers toward testing, agent design and process analysis.
- Measure outcomes, not bots. Track hours saved, error rates and cycle time rather than the number of bots in production.
Example: Bank Reconciliation
| Today with RPA | With AGI | |
|---|---|---|
| New statement format | Each bank’s statement format needs its own extraction logic, debit/credit rules and matching rules. | The system receives a statement from a bank it has never seen, in whatever format or language. It works out which columns hold dates, amounts and references, and how debits and credits are shown. |
| Missing references and unusual statements | Statements with no reference numbers, in Arabic, or in an unusual layout are usually left for a person to handle. | It then decides how to match entries when there is no reference number, using amounts, dates and counter party names. It posts the matches and sends a person only the cases that need real judgment, such as an unexplained large difference. |
| Layout changes | If a bank changes a column position or sign convention, the bot breaks and a developer has to fix it. | When the bank later changes its layout, the system adapts without anyone rewriting code. |
In this example, adding a new bank would no longer mean weeks of design, coding and testing, and the work would shift from building bots to reviewing the system’s decisions.
The takeaway
AGI will not make automation irrelevant; it will make process knowledge, governance and clean integration the scarce things. Teams that treat today’s AI agents as practice for that shift will be ready whenever it arrives, and will have built the controls to use it safely.
