Predictions of enterprise resource planning’s (ERP’s) demise are not new. Client-server was supposed to kill the mainframe, cloud was meant to kill on-premise ERP, and best-of-breed applications were forecast to dismantle the suite. It didn’t happen. Instead, incumbents adapted and survived.
Now comes agentic artificial intelligence (AI), the latest technology to rattle the ERP cage. For some, it’s a natural progression. A chance to add real automation to manage clunky and cumbersome tasks. For others, it’s an opportunity to break the mould and rewrite enterprise software history.
Analysts are divided on where this shift will lead. A recent PwC survey found that 79% of executives say AI agents are already being adopted in their companies, and two-thirds report measurable productivity gains.
Notably, there was a 30% increase in executive trust when agents were used for workflow prioritisation, a sign that confidence is beginning to grow compared with the early days of generative AI. But “workflow prioritisation” is not the same as running payroll or reallocating inventory. Translating rising trust in task management to the high-stakes world of ERP will require a much deeper level of assurance, transparency and governance.
Fresh research from BearingPoint highlights just how early most organisations still are in the adoption curve. In the UK, 17% of organisations are only learning about agentic concepts and 23% have identified it as a future priority. A third (31%) are piloting systems, while just 12% are scaling them enterprise-wide.
Where projects are under way, more than half (51%) anticipate operational efficiency gains, 42% expect new AI-driven services, and another 42% are looking to greater resilience and adaptability. But the timelines are stretched. Only 23% expect full return on investment in two years, while more than a quarter anticipate a wait of five to seven years or more.
Meanwhile, Gartner recently claimed that more than 40% of agentic AI projects will be cancelled by 2027 due to high costs and unclear business value. The firm warns of widespread “agent-washing”, with suppliers rebranding assistants and chatbots as agents without delivering real autonomy. It’s a real concern, and one that muddies the waters for enterprise buyers.
Hype game
Suppliers are certainly moving quickly to showcase their own agentic capabilities. SAP, Oracle, Microsoft, Infor and Unit4 all now talk about “AI copilots” that will evolve into autonomous agents able to take on finance, HR or supply-chain tasks.
Only recently, Oracle unveiled “50+ role-based AI agents” it claims will “fully automate end-to-end business processes” in its Fusion suite. Microsoft describes Dynamics 365 agents as “redefining how business processes are orchestrated and executed”, while SAP has promised Joule Agents that can “autonomously plan and execute multi-step workflows”.
Such marketing highlights the scale of ambition, but it also raises questions about potential hype and overreach. Of course, this is not new to supplier marketing, but history is littered with expensive mistakes, where early adoption did not necessarily convert into the promises emblazoned in glossy brochures.
For this reason, customers are understandably wary. As Nathalie Kuosa, chief financial officer and chief operating officer at Epical Group, a Unit4 customer, points out, her company is only beginning pilot work to see whether AI can help allocate resources more efficiently. She says that it’s “early days”, but that she is clearly hopeful, especially when it comes to removing some manual tasks. Not everyone feels the same.
Dead horse?
Seth Ravin, CEO of Rimini Street, tells Computer Weekly: “It will come to an end over the next decade, because the agentic format is a next-generation paradigm that replaces what we know as ERP today.”
He argues that upgrades and migrations are a “waste of time,” urging companies to bypass expensive projects and instead use small-scale “explorer packs” to automate transactions above existing systems. He points to a pharmaceutical deployment in Brazil where more than 70% of manufacturing processes have already been automated in this way. In his agentic AI vision, ERP does not evolve, it simply disappears, transaction by transaction, until it is no longer needed.
Not everyone agrees. Conor Riordan, chair of the UK and Ireland SAP User Group, insists ERP is far from finished. “ERP is not dying, it’s evolving quickly,” he says. “The emergence of agentic AI doesn’t signify the end of ERP, but rather its transformation. Core ERP foundations will remain intact, but they will be augmented by AI to enhance automation, adaptability and insight. Just as cloud computing revolutionised enterprise software, agentic AI is poised to drive the next significant evolution.”
Riordan also highlights the practical reality facing SAP customers. “With the launch of S/4Hana a decade ago, it is inevitable that ECC support will conclude, meaning customers are under pressure to replatform before the 2027 deadline,” he says. “As agentic AI emerges on SAP’s cloud platform, numerous customers are crafting their upgrade business cases to capitalise on potential innovation, while remaining cautious about additional disruption, costs and complexity.”
For Riordan, customer enthusiasm for agentic AI is real, but so are the concerns – certainly around governance frameworks, transparency in decision-making and cost predictability. One particular risk is what he calls “licensing creep”, as suppliers may treat AI agents as additional users or indirect access, triggering unforeseen costs.
For Ravin, this is clearly a moment for disruption. “Cost to serve is everything,” he says, adding that agentic ERP can lower the cost to serve “without throwing money at migrations that deliver no real value. The migration effort is not worth the money, time and effort,” he insists. “Instead, our position is, save the money from the upgrade. You can leapfrog to agentic AI ERP,” argues Ravin.
In his view, even suppliers know the traditional model is running out of road, pointing to SAP’s Business Technology Platform and “clean core” messaging as evidence that processes are already being lifted above ERP.
Overcoming complexity
Many customers will no doubt be looking at their options. On another front, ERP has always been saddled with a complexity image (regardless of but probably because of the many upgrades, customisations and workarounds on the market). It stands to reason that hope is being put on agentic AI to simplify things and iron out some of those complexities.
The gradual evolution of ERP systems, from using copilots to adopting agentic AI, promises a lot, but as Jeremy Ung, chief technology officer (CTO) of BlackLine, argues, this shift must start with trust. “Agentic AI represents the next step beyond copilots in ERP, moving from suggestion to action,” he says. “In finance and operations, that shift only works if the systems are transparent, auditable and designed with domain knowledge at their core. Businesses already struggle with ERP complexity and compliance, so trust in automated decisions is essential.”
For Ung, the safe entry point lies in high-volume, repetitive finance tasks such as closing books faster, reconciling accounts and surfacing anomalies, areas where automation can free teams for more strategic analysis without undermining governance.
Roman Zednik, EMEA field CTO at Tricentis, agrees. “The promise of agentic AI in ERP is compelling, but the real question is how to deliver it safely, without breaking the trust that underpins enterprise systems,” he says.
If agents are not transparent, governed and rigorously validated, he warns, they risk compounding an already complex IT stack.
Zednik stresses that governance is non-negotiable. “With approaches like agentic test automation, businesses can ensure that AI agents operate with the same discipline as skilled users, continuously testing, validating and flagging risks before changes reach production,” he says.
In Zednik’s view, the winners will be those who deliver autonomy without introducing fresh uncertainty.
Managing expectations
That’s a key point, because we have seen it so many times over the years, where a new technology emerges, solves some old problems, but really only creates new ones in the process. It demands a softly softly approach, as Kuosa at Epical has already suggested.
Ung notes that early adopters are not reinventing ERP wholesale, but targeting narrow, repetitive use cases that can demonstrate value safely. Simon James, managing director for data science and AI at Publicis Sapient, says this incrementalism is the right approach. “If you can’t explain how a system could go wrong and mitigate the reasons, you are not in a position to deploy,” he says.
For James, agentic architectures can actually improve transparency by providing granular logs of every action rather than black-box outputs. He recommends a “last-mile first” strategy, where AI is used to enhance and validate human work before replacing it. That way, adoption builds confidence, creates natural audit trails and generates return on investment in a way that stakeholders can trust.
Others also see agentic AI adoption unfolding in stages. Alfred Obereder, partner at BearingPoint, points to Microsoft’s own mantra. “It’s called Copilot, not autopilot,” he says, setting out a phased path. Copilots act first as personal assistants; then digital colleagues join teams under human supervision; and only later do agents run business processes end-to-end, reporting back to humans as needed.
“Governance should scale with the autonomy of the agent and the potential impact of its actions,” argues Obereder, calling for a “zoned” model that matches oversight to business risk.
Tarang Puranik, executive vice-president at Infosys, takes a similar line. “ERP systems are rich in data and handle vast volumes of rule-based transactions every day,” he says. “Agentic AI takes this a step further by realising the vision of the autonomous enterprise.”
But Puranik adds that responsibility must come first. “Customers want assurances that this shift will not increase risk or complexity,” he says. “Embedding governance frameworks, ensuring explainability of AI decisions and phased adoption are essential.”
Infosys typically advises clients to start with finance process automation under “human oversight” before expanding into more complex areas. So, for now at least, ERP’s obituary looks premature. As Riordan notes, most customers are still working through upgrade pressures while exploring how agentic AI might fit. What’s clear is that enthusiasm for automation is matched by concern over governance, transparency and cost.
That leaves customers with a choice. Take a cautious leap and run small pilots, or hold back and see how the technology fares under scrutiny. For some, the safe option may be to watch what happens in adjacent sectors (finance, HR, supply chain) where agentic AI is already being trialled, and wait for proven outcomes before committing to core ERP processes.
The coming milestones will matter. SAP’s 2027 ECC support deadline will force many to make decisions about upgrades and cloud adoption, while AI regulation in Europe and beyond will bring sharper focus on governance and explainability. These may prove more natural moments for adoption than rushing into large-scale experiments today.
To borrow from Mark Twain, perhaps the stories of ERP’s death are indeed a little exaggerated. But agentic AI has cracked open the debate once again, and this time the stakes feel higher.
