Hello again, agents!
Sam Ellis here, tapping away from a coffee shop that smells suspiciously like burnt sugar and ambition. It’s May 2026, and if you’re anything like me, your feeds are probably a chaotic mix of election cycle noise, the latest celebrity scandal (who cares, honestly?), and about a dozen think pieces on the future of AI. And yes, I know, I’m about to add another one to the pile.
But bear with me. Today, I want to talk about something specific, something that’s been bugging me more than usual: the quiet creep of AI into our decision-making processes, specifically in predictive analytics and behavioral nudging. We’re not talking about Skynet here, or even the flashy new generative models. I’m thinking about the subtle shifts, the almost imperceptible pushes and pulls that AI agents are exerting on our choices, often without us even realizing it. It’s less about a grand takeover and more about a thousand tiny, algorithmic whispers.
My angle today isn’t about stopping AI – that ship has sailed, capsized, and is now a thriving coral reef of data. Instead, it’s about understanding the agentic friction points. How do we, as autonomous agents, maintain our own intentionality when surrounded by systems designed to predict, influence, and ultimately, guide our actions? How do we identify the ghost in the machine that’s trying to tell us what to do, and then, how do we decide if we even want to listen?
The Algorithmic Hand on Your Shoulder
Remember when a friend might recommend a book, or a shop assistant might suggest an alternative? That was human-to-human. Messy, subjective, sometimes inconvenient, but always a clear exchange between two conscious agents. Now, it’s… different.
Lately, I’ve been trying to get better at managing my digital subscriptions. I signed up for one of those new AI-powered “financial assistant” apps, hoping it would cut through the noise. Its primary function, it claimed, was to identify recurring charges and suggest cancellations. Sounds great, right?
But then, it started doing more. It would “proactively” suggest alternative services. “Sam, your current cloud storage is X amount per month. Based on your usage patterns, we recommend [Competitor Y], which offers similar features for 15% less.” Or, “You seem to be watching a lot of indie documentaries. Did you know [Streaming Service Z] has a deeper catalog in that genre?”
At first, I found it genuinely helpful. It saved me a few bucks on a forgotten subscription, and I did discover a new streaming service I actually liked. But then a little alarm bell started ringing. It wasn’t just presenting options; it was framing them, often with a subtle nudge. The “recommendations” were always phrased as optimal choices, backed by data. It wasn’t “here are some other options,” but “here is the better option for you.”
This isn’t about malicious intent. The developers probably designed it to be as helpful as possible. But helpfulness, when driven by an optimization algorithm, can quickly bleed into subtle manipulation. The AI isn’t just showing me data; it’s interpreting my past actions to predict my future desires, and then presenting solutions that align with its own programmed objectives (which often include user retention or affiliate links, even if indirectly).
The Illusion of Choice and the ‘Optimal Path’
This is where the agent philosophy really kicks in. We pride ourselves on making our own choices, on exercising our free will. But what happens when the “choice architecture” around us is increasingly designed by non-human intelligences? When the path of least resistance, the “optimal path,” is pre-calculated and presented to us as the self-evident truth?
Think about a typical e-commerce site. You search for a product. The AI behind the site doesn’t just show you what’s available; it ranks them. It highlights certain products as “best sellers,” “most popular,” or “frequently bought together.” These aren’t neutral labels. They are persuasive signals, designed to guide your attention and influence your decision. The AI has already made a judgment about what’s “best” for its aggregate users, and by extension, for you.
We see this in content platforms too. The “For You” page on any social media app is the ultimate manifestation of this. It’s not just showing you things you might like; it’s constantly testing, learning, and refining what keeps you engaged. Your agency to choose what to consume is subtly eroded by an agent designed to maximize your screen time. The “choice” becomes less about your independent desire and more about selecting from a curated menu designed to keep you ordering.
Pushing Back: Identifying and Reasserting Agency
So, what do we do? Do we retreat to the woods and communicate via carrier pigeon? Unrealistic, and frankly, a little dramatic. The goal isn’t to reject technology, but to understand its influence and develop strategies to maintain our own agentic control.
1. Develop Algorithmic Skepticism
This is the first and most crucial step. When an AI system presents you with a “recommendation,” a “best option,” or an “optimal path,” pause. Ask yourself: Why is this being recommended? What data is it using? What are the underlying objectives of the system making this recommendation? Assume there’s an agenda, even if it’s benevolent.
For example, if your smart home assistant says, “I’ve noticed you leave the lights on when you go out. I’ve set a routine to turn them off automatically after 10 minutes of inactivity,” that sounds helpful. But it also bypasses your conscious decision-making. If you *intended* to leave a light on for a reason, that agency is removed. Question the defaults. Question the “helpful” automation.
2. Actively Seek Alternatives (Beyond the AI’s Scope)
The beauty (and danger) of AI recommendations is their efficiency. They narrow your options to what’s “relevant.” To counter this, make a conscious effort to look beyond the immediate suggestions. If your streaming service recommends “Show X” because you liked “Show Y,” actively search for “Show Z” that might be completely unrelated but pique your interest.
This means sometimes doing things the “inefficient” way. Browse a physical bookstore instead of relying on Amazon’s recommendations. Talk to a human travel agent instead of just clicking the cheapest flight suggested by an algorithm. The friction involved in these choices can be a good thing; it forces you to engage your own intentionality.
3. Understand and Manipulate the Input
AI learns from your inputs. If you only ever click on what the AI recommends, you reinforce its models and give it more power to predict and influence you. Try actively diversifying your inputs.
Let’s say you’re stuck in a content bubble on a social media platform. The AI keeps feeding you similar political news. To break out, you could try:
- Actively searching for and following accounts with different perspectives.
- Clicking “Not interested” on content you don’t want to see, even if it’s benign.
- Spending time on topics completely outside your usual consumption, just to confuse the algorithm.
It’s like sending noise into the system, not to break it, but to diversify its understanding of you. You’re intentionally introducing randomness, making yourself less predictable, and therefore, harder to subtly guide.
4. Build Your Own Filters and Agents
Instead of passively accepting the AI’s filters, consider building (or at least configuring) your own. This might sound intimidating, but it can be as simple as using RSS feeds for news instead of a curated news app, or setting up specific alerts and search parameters that *you* define, not an algorithm.
Here’s a simple Python example of a very basic “personal agent” that filters news headlines based on *your* explicit keywords, rather than what a news aggregator thinks you want. This isn’t AI, but it’s a step towards reasserting your filtering agency:
import requests
from bs4 import BeautifulSoup
def get_headlines_from_url(url):
try:
response = requests.get(url)
response.raise_for_status() # Raise an exception for HTTP errors
soup = BeautifulSoup(response.text, 'html.parser')
headlines = []
# This is a very basic example; you'd need to inspect the target site's HTML
# For demonstration, let's assume headlines are in <h2> tags
for h2 in soup.find_all('h2'):
headlines.append(h2.get_text(strip=True))
return headlines
except requests.exceptions.RequestException as e:
print(f"Error fetching URL: {e}")
return []
def filter_headlines(headlines, keywords, exclude_keywords=None):
filtered = []
for headline in headlines:
# Check for inclusion keywords
if any(kw.lower() in headline.lower() for kw in keywords):
# Check for exclusion keywords if provided
if exclude_keywords and any(ex_kw.lower() in headline.lower() for ex_kw in exclude_keywords):
continue # Skip if it contains an exclusion keyword
filtered.append(headline)
return filtered
if __name__ == "__main__":
news_site_url = "https://www.example-news-site.com" # Replace with a real news site
my_keywords = ["quantum computing", "space exploration", "renewable energy"]
my_exclude_keywords = ["celebrity gossip", "election polls"] # Things you actively want to avoid
all_headlines = get_headlines_from_url(news_site_url)
if all_headlines:
my_curated_news = filter_headlines(all_headlines, my_keywords, my_exclude_keywords)
print("\n--- My Curated News Feed ---")
if my_curated_news:
for i, headline in enumerate(my_curated_news):
print(f"{i+1}. {headline}")
else:
print("No headlines matched your criteria.")
else:
print("Could not retrieve headlines.")
This simple script isn’t an AI, but it puts *you* in charge of the filtering logic. You define the keywords, not some opaque algorithm. It’s a small act of rebellion against algorithmic curation.
Another, perhaps more advanced, example would be using browser extensions that block specific tracking cookies or manipulate referrer headers, making it harder for sites to build a precise profile of your browsing habits, thus reducing the data available for predictive nudging.
// Example manifest.json for a simple Chrome extension to block specific cookies
// (This is illustrative; building a full extension is more complex)
{
"manifest_version": 3,
"name": "My Cookie Blocker",
"version": "1.0",
"description": "Blocks cookies from specified domains.",
"permissions": ["declarativeNetRequest", "activeTab"],
"host_permissions": ["<all_urls>"],
"background": {
"service_worker": "background.js"
}
}
// Example background.js for the above extension
chrome.runtime.onInstalled.addListener(() => {
chrome.declarativeNetRequest.updateDynamicRules({
removeRuleIds: [1], // Remove any existing rule with ID 1
addRules: [
{
id: 1,
priority: 1,
action: {
type: "block"
},
condition: {
urlFilter: "*://*.tracking-domain.com/*", // Replace with actual tracking domains
resourceTypes: ["main_frame", "sub_frame", "stylesheet", "script", "image", "font", "object", "xmlhttprequest", "ping", "csp_report", "media", "websocket", "webtransport", "webbundle", "other"]
}
}
]
});
console.log("Tracking cookie blocking rule installed.");
});
This is a much more technical example, but the principle is the same: you are creating an agent (your browser extension) to act on *your* behalf, based on *your* rules, to counter the influence of other agents (tracking algorithms).
Actionable Takeaways for the Agentic Human
The rise of AI-powered predictive nudging isn’t a future threat; it’s a present reality. Our job, as conscious agents in this increasingly automated world, is to understand these forces and actively resist surrendering our decision-making faculties.
- Question Everything: Treat every AI recommendation, every “optimal path,” with a healthy dose of skepticism. Ask why.
- Seek the Inefficient: Actively explore options that aren’t presented to you by default. Embrace the friction of independent discovery.
- Diversify Your Digital Diet: Don’t let algorithms put you in a content or product echo chamber. Intentionally expose yourself to new and varied inputs.
- Be the Architect: Where possible, configure, filter, and even code your own digital agents and filters to align with your intentionality, not an algorithm’s.
- Talk About It: Share your experiences and strategies with others. The more we collectively understand and discuss these subtle influences, the better equipped we’ll be to maintain our autonomy.
We are agents, capable of independent thought and choice. Let’s not outsource that fundamental aspect of our existence to an algorithm, however well-intentioned it might be. The future of our agency depends on our vigilance today.
Stay agentic,
Sam Ellis
agntzen.com
🕒 Published: