The agentic AI information asymmetry problem
“The future is already here — it’s just not evenly distributed.” — William Gibson
Scrolling through my online feeds and developer forums it’s abundantly clear there is a huge divide emerging around developer knowledge of AI. Many developers dismiss AI, commonly with complaints of ‘AI slop’ or ‘AI creates tech-debt’. While a smaller minority claim benefits as profound as 5-10x productivity gains (or more), and are already planning to upend how they build their companies.
So, what’s going on here? Is this just the typical ‘hype cycle’ playing out for AI? I’m here to tell you that it most certainly is not. It’s a much more insidious, information asymmetry.
Note: If this all feels like ‘too much hype’ , simply skip to the end for practical advice on how to ‘accelerate yourself to your ‘aha’ moment’. Don’t just take my word or anyone else’s,. close your browser and try the tools for yourself.
What’s causing AI information asymmetry?
The rapid evolution of models and agentic software development
The speed of which AI is developing is profound. Models are improving, but so are techniques of agents. Agents are (in a primitive sense) LLMs that run in loops, where the LLM receives feedback and then makes adjustments accordingly.
Agents allow a degree of accuracy and autonomy that far exceeds the capabilities of code assistants like early GitHub Copilot. Opinions formed even a few months ago are now effectively obsolete. Are we talking about agentic software development or simply code assistants? Because the productivity differences are truly profound.
Asymmetries created across disciplines like front end and infrastructure
The AI revolution is affecting some areas of development more than others. Front-end developers using React have seen incredible advancements with tools like V0 and Lovable. While infrastructure engineers working with tools like Terraform are having far less fruitful experiences. The disparity between ecosystems comes from the fact that:
- Models are not trained on all programming languages and ecosystems equally
Feedback loops are harder in ‘systems environments’ like complex backend infrastructure, microservices architectures and infrastructure.
Front end is visual, so AI tools like Loveable and V0 are more effective in virality
It’s certainly possible we see monolithic architectures making a comeback or certainly code structures and design patterns that are more designed AI first.
High bandwidth knowledge sharing in tech hubs like San Francisco
Next it’s hard to say exactly how much location affects the spread of information, but for sure the high concentration of innovation happening in tech hubs like San Francisco or New York create feedback loops of adoption and skill development that simply aren’t matched in other regions. Online meetups, events, papers and the like are helping to disseminate information across borders and oceans, but there is no real substitute for the high bandwidth of synchronous learning and communication.
Developer patience prevents getting to the ‘aha moment’
Many developers are quick to dismiss AI tools as ‘not ready for prime time’ or ‘not as good as a human developer’. This comes from subconscious bias and the fact that developers are inherently skeptical of shiny technology waves. But AI is different, it’s like the internet, cloud, and mobile combined, rather than passing waves like blockchain.
Many experienced developers will try AI tools briefly, provide minimal context to the tool, and then dismiss it when the tool doesn’t magically read their mind. AI needs to be treated like a junior developer with total amnesia. You must effectively preload context with effective prompt engineering for maximum effect. One day, the tools will work with next to no configuration, but at least for now you must manually put in some work.
Closing the asymmetry gap
What I find personally challenging is that I can’t simply explain away the asymmetry. Reading this article alone won’t convince you. These tools have to be experienced firsthand. The good news is that getting to aha can take as little as one hour of intentional learning.
The fastest way I know is to:
Download an AI editor like Cursor (Cursor works with Gitpod), Windsurf (Or use a CLI or extension like Cline or Claude Code)
Set it in ‘agent mode’ where it will get feedback from command execution
Spend enough time exploring, giving context etc to really witness the power
This last point is crucial. You must spend enough time exploring the tools, giving them context, and prompting them appropriately. If you’re just assessing autocomplete functionality (which still exists and is useful) you’re not seeing the bleeding edge of what’s possible with AI tools that now run autonomously. The feedback loop created when the AI can run tests, build code, and analyze logs enables it to solve problems of far greater complexity than before.
This next weekend might be very consequential
I hope this article helps you understand why some of the discourse online feels so disjointed. If you’ve not yet experienced the power of agentic AI tools, I encourage you to try them out for yourself. If you have already reached the aha moment, congratulations. Now go share your workflows with colleagues (not just your conclusions!). Help them understand that patience and proper context setting are required.
If you’re an employer, make space for your employees to experiment with these tools. If developers are stuck on the daily grind, the ‘one step back’ to sharpen your saw with AI could see you take ten steps forward every day from now on.
And finally, consider this weekend an investment in your future: schedule a few hours to explore AI tooling. The productivity gains and competitive advantage you’ll gain will pay dividends for years to come—an opportunity that you can’t afford to miss. As Ravi Gupta quotes the late and great Aryton Senna in his article AI or Die (and it’s worth reflecting on): “You cannot overtake fifteen cars in sunny weather. But you can when it’s raining.”
If you’re past that aha moment with AI check out the Gitpod Agent SDK. With secure agentic enabled environments you can do seamless agent to developer handovers in isolated and secure Development Containers. Editors like Cursor come as standard.
We’re bringing vibe coding to the enterprise.
