Published on the 22/04/2026 | Written by Heather Wright
Early gains meet integration and infrastructure costs…
Agentic AI is rapidly moving from promise to practice for software development and data teams, but two new reports suggest the biggest challenge isn’t whether the technology works, it’s whether organisations can afford to wire it into everything else.
MIT Technology Review Insights’ report, Redefining the Future of Software Engineering, shows a surge of confidence that agentic AI – the advanced, autonomous AI systems capable of independently executing multi-step workflows – will dramatically accelerate software development. But DBT Labs’ 2026 State of Analytics Engineering Report, which tracks how AI is reshaping analytics and data platforms, highlights a quieter, but consequential trend: Infrastructure and integration costs are rising faster than budgets, particularly as AI becomes embedded across analytics and engineering workflows.
“This raises a central question for 2026: Whether validation, testing and oversight mechanisms are scaling at the same pace as AI-driven output.”
Together, the reports highlight that AI agents might be here, with adoption is accelerating, but the bill is landing not with licenses but with compute, orchestration and governance.
MIT Technology Review Insights surveyed 300 senior IT, software engineering and data leaders at large enterprises and found that 98 percent expect agentic AI to accelerate software delivery, with average speed gains of 37 percent anticipated over the next two years. More than half already rank agentic AI among their top three software investments, with that figure expected to rise to 84 percent within three years.
“2025 has marked a steep acceleration in how AI is used in software engineering, evolving from vibe coding experiments to more disciplined agentic engineering,” the report notes. Productivity no longer scales linearly with headcount. “It scales with how clearly we define the product vision, engineer context and onboard agents.”
Ambitions, too, are scaling quickly, escalating from assistance to autonomy, with agents managing product development and software development lifecycles. That’s a goal 72 percent aim to achieve in two years and 41 percent expect within 18 months.
The DBT Labs report shows the same momentum in analytics engineering. AI-assisted coding is now prioritised by 72 percent of analytics teams, and among leaders more than 77 percent emphasised AI for productivity gains.
“AI is no longer experimental inside data teams,” the report says. “Instead, it’s funded, embedded and actively reshaping analytics engineering workflows, influencing how code is written, how insights are generated and how teams invest in analytics and data infrastructure. What was once exploratory is now operational.”
Despite the optimism, the MIT report injects a note of caution: Most organisations don’t expect dramatic gains in the short term. The majority of respondents expect slight (14 percent) or at best moderate (52 percent) improvements from agentic AI over the next two years. Only nine percent expect the impact to be game-changing in that timeframe.
That detail reframes the adoption curve: Organisations are absorbing the full cost and complexity of integration before seeing transformational returns.
The DBT report reinforces this. AI’s primary impact today is operational – shortening cycle times, increasing throughput and expanding output volumes – rather than fundamentally changing what teams deliver.
The hidden cost of integration
The two reports align when it comes to the hidden costs, and a key challenge, involved.
In the MIT study, agent integration with existing enterprise applications and the ongoing cost of computing resources are tied as the biggest challenges to scaling agentic AI, each cited by 44 percent of respondents. Multi-agent orchestration across complex enterprise environments, often involving hundreds of APIs and third-party and home-grown applications, is a major constraint. Orchestration agents exist, but are not yet able to manage lifecycles end to end, partly because of that complexity.
The DBT Labs report shows the same pressure playing out in the analytics sphere. Fifty-seven percent of the 363 data practitioners and leaders surveyed reported increased data warehouse and compute spend as AI-drive workloads expand. Infrastructure demand is absorbing much of the AI investment, the report notes. Meanwhile, only 36 percent report increased team budgets.
While AI tools may be inexpensive to access, the reports point to the expense of operating AI at scale. Compute, storage, orchestration and observability costs quickly accumulate, especially when AI expands workload intensity by default.
Both reports also flag a growing imbalance between output and control.
DBT Labs notes that while 72 percent of teams prioritise AI-assisted coding, only 24 percent prioritise AI-assisted pipeline management, including testing, validation, observability and quality controls. “The signal is clear: Teams are leveraging AI to accelerate creation, but not to reinforce governance at the same pace.”
MIT’s research echoes this concern from a software engineering perspective. Reliability, governance and auditability are cited as gating factors to wider deployment, particularly in regulated or safety-critical environments. Most agentic systems today still require human oversight, which limits how quickly autonomy can scale.
DBT’s report found that trust in data remains a foundational concern as AI use expands. Poorly governed data doesn’t just undermine outcomes, it increases operational risk, forcing organisations to slow deployment or add safeguards after the fact.
“As AI expands what analytics teams can produce, concern is rising alongside adoption. This raises a central governance question for 2026: Whether validation, testing and oversight mechanisms are scaling at the same pace as AI-driven output.”
Or as one data engineer quoted in the report notes: “AI won’t fix a messy foundation. It just makes the lack of discipline much more visible.”
