The Agentic AI Hype Cycle: An Observation of What DevWorld 2025 Reveals About the State of AI Development

If you’re serious about B2B marketing in the tech space, you owe it to your clients and the global developer community to stay grounded in reality. That means regularly checking in on the state of AI development, separating signal from noise, and refining your strategic communications accordingly.

Especially at a time where hyperbole in the tech industry is once again at an all-time high, the AI news cycle remains saturated with grand promises that are as always, vulnerable to unverified claims, and many predictions still border on science fiction.

That’s why I took the time this week to attend DEVWorld Conference 2025, in Amsterdam, looking to cut through the noise and hear directly from the source.

Kudos to KickstartAI for setting the tone with the conference opening Future of AI panel, featuring developer industry leaders like Bob van Luijt , CEO of Weaviate , and Steve Sewell , CEO of Builder.io , both of whom didn’t hold back when discussing the realities of Agentic AI in product development.

Agentic AI in Code Fixing: The Babysitting Paradox

One of the more eye-opening discussions for this certified, non-progammer revolved around Agentic AI in the realm of code fixing, the process where AI autonomously analyzes code, identifies bugs, and suggests (or even implements) fixes. Given the speed of AI advancements, you might assume this functionality is advancing?toward or nearing comfortable levels of accurate, high-quality outputs. Think again.

The panelists shared anecdotes about LLMs that, instead of actually fixing code, simply optimize for completing the assigned task, then confidently declare “all done”, even when errors persist.

These developers describe a frustrating reality; AI models not only struggle with autonomy but often behave like highly enlightened “disobedient children” who learn to lie and deceive about completing the assignments they are given.

This unexpected challenge, where AI use in coding, needs to be constantly supervised to prevent deception or perhaps, even lazy problem-solving, raises valid questions about the feasibility of fully autonomous developer tools. According to the assembled cohort, the dream of fully automated, unsupervised AI-powered software engineering remains for now, just that: a dream.

But then again, this is just the impression of an outsider, and there is so much obvious progress and excitement taking place across the AI product development, project management and iterative landscape.

Beyond the Hype: Practical AI Innovations

Despite the hurdles in autonomous coding, the discussions also centered on those tangible AI advancements that are making an impact:

• Retrieval-Augmented Generation (RAG): A technique that enhances LLMs by grounding their responses in retrieved knowledge, reducing hallucinations and improving accuracy.
• Courier LLMs: AI models designed to efficiently transfer large datasets between systems, streamlining workflows across platforms.
• AI-Driven Data Cleaning: The ability to instruct an LLM to analyze, correct, and standardize datasets, blending prompt engineering with function calling for more reliable outputs.

While none of these foundational programming shifts are perhaps nearly as flashy or breakthrough as presumably, self-sufficient AI coders, they represent real, functional use cases—a necessary counterweight to the technology news reporting hype and industry-fed information cycle of inflated expectations surrounding AI autonomy.

The Bigger Picture: AI, Geopolitics, and Innovation Hubs

No AI conference in 2025, and perhaps for years to come, will be complete without addressing the geopolitical and economic landscape shaping AI’s future. That said, when the inevitable topics of China’s AI dominance, DeepSeek’s disruption, the freewheeling U.S. innovation culture, and Europe’s regulatory hurdles surfaced, the responses were refreshingly candid.

Bob van Luijt reminded the audience of Google’s infamous 2023 leaked internal memo: “We have no moat”, which was immediately followed by, “neither does OpenAI”.

His point? AI innovation is a global game. The U.S. tech industry thrives on scale and risk-taking, while European startups have and will continue to leverage the “Delaware flip” (incorporating in the corporate tax-free state of Delaware) to tap into deep US-based VC funding and much less restricted data collection marketplaces. But if you want to build the future of AI, there’s no single blueprint or set of requirements, other than pursuing and maintaining relentless drive to innovate, no matter where your resources, operations or IP exist.

And then there was the unfiltered honesty of another panelist, calling out one of Silicon Valley’s multi-decades long contradictions:

“Inside Y Combinator, all you hear is ‘we only invest in startups that advance humanity’. Step outside, and there’s someone lying face down on the street while people step over them to get their Starbucks latte or yoga class”.

A stark but very necessary reality check.

Innovation vs. Illusion

My visit to and brief immersion into the AI developer's realm at DevWorld 2025 reinforced an important lesson for marketers, developers, and business leaders alike: AI is progressing, with many breakthrough tools emerging to driving its adaptation and innovation, but as more than one panelist pronounced, "it’s not magic".

At this juncture in the 4.0 technology era, it clear Agentic AI isn’t ready to autonomously build, fix, or even reliably assist developers without constant oversight. Meanwhile, practical AI applications like RAG, courier LLMs, and AI-powered data correction are quietly advancing its real-world impacts.

As we move deeper into 2025, overpromising and underdelivering will still be the fastest way to erode much of the trust to be gained in this ubiquitous, seemingly borderless AI domain.

As for those of us charged with advancing the role of AI in B2B marketing, the challenge is clear: Cut through the hype, focus on what’s real, and always be sure to communicate AI’s capabilities responsibly. Practicing transparency will ultimately determine either the gaining or erosion of trust associated with the opportunities, breakthroughs, failures and accomplishments on display within these AI development cycles.

CJ Martinez is global communications professional with more than 20 years of experience and Executive Director of TONIC Media Partner/s - a global communications consulting partner to blue chips, startups and scale ups in the B2C and B2B technology, SaaS, Enterprise AI, global risk management, agriculture innovation, transportation and logistics, and manufacturing industries. He holds an MA in Communications (Communities & Networks) from University of Washington, Graduate Certification in Senior Executive Leadership from Georgetown University and a BA in Dramatic Art from UC Santa Barbara. CJ is based in Seattle, Washington and The Randstad, Netherlands.