Baris Kavakli (Portera) discusses his predictions for agentic AI in 2026
As agentic AI tools continue to transform enterprise software, organizations that stay ahead of the curve will be among the winners. We spoke with Baris Kavakli, managing partner of Amsterdam-based technology consultancy Portera, who shared a few predictions on what can be expected in 2026.
This year will mark a pivotal turning point for AI in the corporate world. There has been massive investment in AI in recent years, but the organizations that emerge successful will be those that plan best, not the ones that spend the most.
This transition represents the fastest adoption of technology in history and it introduces a significant set of challenges for organizations. Kavakli shared a number of predictions on where agentic AI is heading in 2026.
Winners and losers
Firstly, he believes that agentic AI will go truly mainstream, but that reality will set in for many organizations that were not ready for it. The reality is that many projects will ultimately fail.
“The shift from generative AI to agentic AI represents a fundamental change in how enterprises think about artificial intelligence,” says Kavakli.
“Think of it as the journey from infancy to adolescence. Generative AI was the infant stage – remarkable capabilities, but entirely dependent on human prompting. Agentic AI is the teenager: capable of independent action, but requiring guardrails, supervision, and the wisdom to know when to ask for help.”
Kavakli predicts that agentic AI will be a part of most enterprise applications by the end of this year. Whether Salesforce’s Agentforce, Microsoft’s Copilot agents, or one of dozens of other options, assimilating these tools into an organization’s existing structures is easier said than done.
“The pattern I see across the industry is predictable: Impressive demo, enthusiastic sponsor, then slow death by a thousand integration and adoption challenges. The AI tool works beautifully in isolation. It fails spectacularly when it meets the messy reality of enterprise data, legacy systems, and organizational politics,” notes Kavakli.
Organizations that end up being successful with agentic AI adoption will need to have a good data foundation, a good governance framework, clear business outcomes in mind at the start, and will need to treat the human-AI interface as a central design challenge, not an afterthought.
The EU AI Act
Besides the continued development of AI technologies, this year will see another major development: The EU AI Act will become fully enforceable on 2 August, 2026. That means the most comprehensive AI regulations in the world will apply to any company in the EU that uses AI tools.
Indeed, this is huge for European organizations and organizations that operate in the EU. Fines can reach as high as €35 million (or 7% of global annual turnover, whichever is higher).
“The competitive advantage goes to those who embrace compliance-first design," Kavakli says. “Instead of viewing the AI Act as a regulatory burden, leading organizations are using it as a forcing function for good AI governance. If you haven’t started your EU AI Act compliance program, start today. If you have, double your timeline estimates.”
The models are commodotizing
Another prediction from Kavakli is that 2026 will be the year where it no longer is about the models, but the knowledge base and the company ontology that is built around the models. “The assumption that bigger models equal better performance is crumbling under economic reality,” he notes.
Cost is a major consideration here. Spending fortunes on large model APIs might work for some large multinationals, but optimized smaller models are often able to get the same job done just as well and at a fraction of the cost.
Besides that, many companies will find that fine-tuned small models actually work better for their specific needs than a general-purpose large model. These smaller models can be trained in a very specific domain. They do not need to know everything – they are specialized in the domain of that particular business.
“I expect 2026 to reveal a barbell strategy: Giant models for complex reasoning and creative tasks, small models for high-volume operational applications. The organizations that insist on deploying GPT-5-class capabilities for every use case will find themselves out-competed by rivals who match model size to task requirements.”
Skills paradox
Another prediction has to do with a looming skills paradox: While general skills are quickly becoming obsolete, specific AI skills are just as quickly evolving and in higher demand than ever. This goes well beyond simply known how to use commercial LLMs – it entails in-depth knowledge of the limits of AI tools.
“The skills gap isn’t about technology literacy anymore – it’s about human-AI collaboration fluency. Don’t forget: we are the last generation that has managed human-only teams,” notes Kavakli.
When AI is adopted in an organization with employees that are not adequately trained, it can lead to an over-reliance on tools that sometimes produce errors that need to be caught by human review.
Shift in strategy for 2026
Following these predictions, Kavakli recommends organizations conduct a comprehensive AI governance audit, launch a compliance assessment to stay ahead of the EU AI Act, and to be realistic about timelines for AI value realization.
Furthermore, organizations should focus more on scaling AI experiments that work rather than launching new pilots. “Find one area that you believe you can make an impact by moving the needle, and focus there until you get it done.”
“2026 will be remembered as the year AI stopped being something enterprises experimented with and became something enterprises depended on,” Kavakli concludes.
“The organizations that thrive won’t be the ones with the most aggressive AI roadmaps. They’ll be the ones who balance ambition with discipline, who understand that the massive cancellation wave isn’t a threat to be ignored but a warning to be heeded. Agentic AI changes everything. But only for organizations prepared to embrace it.”
