The State of Generative AI in the Enterprise report from Menlo Ventures found that companies are spending more on generative AI. While in 2024, enterprises spent $11.5 billion, this year, companies will more than tripled their AI spending, reaching $37 billion.
Besides an increase in AI budgets, companies are shifting from developing in-house models to outsourcing them from third-party providers in big numbers. The report found that while in 2024, 47% of AI solutions were built internally, and 53% purchased, today, 76% of all AI is purchased rather than developed in-house.
“[Companies are] outsourcing … and just deploying third-party AI, and have no idea absolutely what their AI is doing,” said Drew Naukam, CEO of Gorilla Logic.
“They might be verifying if the AI is working correctly, but don’t really have an idea of how the internal processes work,” said Naukam. “This can lead to a lack of optimization of the models and the benefits of knowing what an AI is doing, how it works within a workflow.”
Companies that outsource all of their AI operations to third-party platforms and don’t build organizational skills internally are at the mercy of vendors and will not reap the benefits that the AI digital transformation can provide, said Naukam.
AI in the workflows: Rethinking linear software development
While Stanford’s 2026 expert predictions claim that 2026 will be the year of AI evaluation,
a highly cited mid-2024 MIT study pinpoints where AI is failing — in the workflows. The MIT report found that 95% of organizations are getting zero return on AI deployed, with most failures found due to “brittle workflows.”
AI can drive efficiency across teams and workflows, not just efficiencies at the individual productivity layer, said Naukam. The challenge is, according to Naukam, that organizations need to have a clear understanding of the workflow they are using before they can even deploy AI.
“You have to have people that have the ability to rethink the workflow at a scale that AI can execute, versus at a scale that humans can execute,” said Naukam.
Some of the differences between AI and humans, for example, are the ability to process multiple different data sets in real time, and the capability to run rapid multiple iterations of a test in real time. These differences are specifically challenging in the software development industry, where workflows are designed to be linear, but AI is non-deterministic. This means that the traditional DevOps workflow—code, test, release—needs to be reimagined for AI iterations.
“You have a problem, and you have the AI interact on that problem, and you try and you fail, and you iterate and you improve, and you repeat that process over and over again to get an outcome that you expect with validation that allows you to validate statistically that the outcomes that you expect are the outcomes you achieve,” said Naukum.
“That’s a non-deterministic process to solve a problem, and that is inherently different from the way Software Engineering has historically functioned,” he added.
Simple workflows can be automated in days and weeks, while more complex workflows may take up to a year to automate, depending on the complexity.
Decision Intelligence: AI at the strategic level
Companies leading in AI are not just deploying AI in production workflows, but across all operations, with strategic departments becoming a new focus.
An International Data Corporation (IDC) white paper found that 88.3% of organizations are using AI at the strategic level, with 88.3% of them having either implemented or planning to pilot Decision Intelligence (DI) initiatives.
“In high-stakes business strategy, the cost of a wrong turn is measured in millions, many of which are irreversible,” Arda Ecevit, co-founder and CEO of NexStrat.AI , an AI management consultant, said.
Leaders can no longer risk a “black box” approach because strategic decisions require accountability and a clear audit trail, he added. “If you cannot see the ‘why’ behind an AI’s recommendation, you can neither commit to the responsibility of acting on it nor defend that decision to a board or stakeholders,” said Ecevit.
NexStrat developed an agentic AI workflow specifically for strategy. The model uses a hypothesis-driven approach, generating strategic hypotheses and then rapidly iterating on them. The iterations and “heavy-lifting” that the AI agent does is traceable across all sources. “This allows leaders to provide high-level judgment and guidance where it matters most,” said Ecevi.
Despite the benefits that AI can bring to the table at the strategic level—enhancing processes, running simultaneous simulations and iterations, and speeding up analysis and generation of content—humans in the loop are necessary, Evici explained.
“AI lacks the ‘last mile’ of human judgment—the vision, intuition, and instinct that define a great leader,” he said. Ultimately, the final accountability and responsibility reside with the human decision-makers, who must remain in the driver’s seat to ensure the strategy is not just analytically sound, but culturally and ethically right.
Disclosure: This article mentions a client of an Espacio portfolio company.
