AI Is Transforming Economics But Enterprise IT Architecture Issues Are Still Here
Investor anxiety around artificial intelligence surged this month in the wake of Anthropic’s announcement that its Claude platform had advanced from a conversational large language model to a task-executing system capable of doing legal work such as contract review and policy drafting. The announcement caused stocks to tumble as investors worried the end was nigh for traditional software and IT services vendors.
See Also: On-Demand | NYDFS MFA Compliance: Real-World Solutions for Financial Institutions
But CIOs shouldn’t start tearing up their vendor SLAs anytime soon.
While AI systems such as Claude lower the marginal cost of writing code and automating discrete tasks, especially when it comes to early-stage work including prototyping and front-end design, the idea that AI will lay waste to the industry is overblown, analysts say.
Enterprises will still need robust architecture, integration and human governance. Tech leaders may find that AI can help fill in gaps in their current developer workforces.
Legacy SaaS vendors whose products already provide scalability, security and integration will have an advantage over companies like Anthropic, whose AI tools have yet to prove they’re enterprise-grade, said Chris Gardner, vice president and principal Analyst, Forrester.
So far AI is driving down cost and helping with prototyping, user-interface generation and knowledge work automation, but for more robust projects, it’s not ready for prime time, he said.
“In some situations, you can create software that’s pretty close to enterprise class,” he said. “But it is a prototype. It is not meant to scale. It may not necessarily be secure.”
Easier lifts such as front-end web development tasks where best practices are well-established will be the first disrupted by AI, but back-end development and integration will continue to be human-led endeavors, especially for enterprises managing many legacy systems.
Core systems still depend on fragmented data architectures, legacy platforms specialized compliance requirements and non-standard integrations that AI alone cannot resolve.
“The process of actually making the connections afterwards is very difficult,” Gardner said. “A human usually has to step in for that.”
AI Adoption: A CIO’s Perspective
But there are some areas in which AI tools can assist in maintaining and modernizing legacy applications, especially those written in older languages.
“We have legacy applications, some built with programming languages that are no longer widely used, which makes finding resources with those skills more challenging. AI has significantly accelerated our efforts to modernize these systems,” said Teena Piccione, secretary and state chief information officer, North Carolina.
Vibe coding has been especially useful for the state when creating prototypes, she said. “AI generated 2,200 lines of code in just 30 minutes for one of four prototypes we submitted to [an] agency to evaluate the best solution for their needs. This capability is helping us mitigate the impact of workforce shortages,” she said.
While there’s a place for vibe coding in software development, it won’t eliminate the need for trained practitioners, Gardner said. In some situations vibe coding can save time and money, but unchecked, AI’s nondeterministic nature could create security and compliance problems.
“Vibe coding is a terrible term,” Gardner said. “Nobody wants a vibe-coded enterprise class app. That’s an oxymoron.” He said a better way to think about leveraging AI is through “vibe engineering,” where AI operates within predefined architectural patterns, security and compliance guardrails. This approach creates composable, consistent systems where humans are still reviewing code.
‘Don’t Fire Your Developers’
Developers most in demand will be those with greater architecture literacy, and roles focused on narrow coding tasks are expected to vanish.
“Don’t fire your developers just yet,” Gardner said.
Industry groups such as the National Association of Software and Service Companies in India agree. “Creating real business value from AI requires careful coordination, with humans in the loop who understand business context, industry knowledge and enterprise workflows,” NASSCOM said.
On the vendor side, cheaper code is leading to pricing pressures, but enterprises remain dependent on those who have deep system knowledge and who move toward offering products with agent-based platforms and interoperability. “If you don’t have a road map for providing agents on your own platform … you’re going to fall behind,” Gardner said.
Piccione expects software vendors to incorporate AI into their ecosystems. At the end of the day, those who get North Carolina state contracts will be the ones providing tangible value, she said.
“Current software and service vendors need to find ways to add distinct value, particularly human value, beyond what emerging technologies can easily deliver,” she said. “From conversations with vendors, it’s clear that many are still trying to sell products or services that AI now generates effortlessly, without offering additional benefits. These vendors must upgrade their sales teams’ skills to be AI literate and highlight the unique value they can provide – value that our organization cannot achieve independently with AI.”
A Place at the Table
IT services firms can also expect a place at the table in the age of AI.
“While the growing excitement around AI platforms and plug-ins may lower the entry barrier for building software, it does not reduce the role of IT services firms,” said Ashok Soota, chairman and chief mentor of IT services firm Happiest Minds Technologies. “Instead, it increases the need for partners that can handle enterprise-grade integration, governance, orchestration and managed services.”
“The idea that new AI tools can simply be plugged into enterprise environments to replace large parts of IT services work is ‘misplaced,'” said Cognizant Technology Solutions CEO Ravi Kumar S. “Enterprises need an understanding of systems and processes that AI tools on their own cannot provide.”