Trust is the constraint
Last week we flagged agent accountability as an open thread—systems breaking down, code ownership evaporating, incident response turning into a game of telephone. This week the thread pulls taut: trust isn't just an agent problem or an ops problem. It's the shared constraint breaking growth tactics, collapsing distribution channels, and stalling every AI feature you're about to ship.
Here's the counterintuitive part everyone's missing. The growth tactic we've spent fifteen years optimizing—referral programs—can systematically attract problematic users. Not because the mechanic is broken, but because growth tactics optimized for velocity now select against the one thing that actually predicts retention: whether someone trusts your product enough to stay. Meanwhile, according to research published in California Management Review, zero-click searches reportedly hit 65% of all traffic by late 2025, and companies are pouring budget into AI features that users will never see because agents don't scroll, don't click ads, and represent the ultimate form of banner blindness. The tools got cheaper. The trust got expensive. And nobody's instrumentation is set up to catch it.
The moment usually arrives quietly. A founder watches their referral program start working—traffic spikes, signups double, the growth curve finally bends upward. Then two weeks later, retention collapses. The new cohort churns at three times the baseline rate. The first instinct is to blame onboarding, or messaging, or some technical friction in the activation flow.
The economics have inverted. For a decade, the constraint was distribution—getting in front of enough people to find your audience. Now distribution is cheap and trust is expensive, because AI has flooded every channel with noise while simultaneously making it trivial to clone any product overnight. SEO used to require months of content strategy; now anyone can generate dozens of articles with a few prompts. Paid search costs have skyrocketed as every company bids on the same shrinking pool of clicks. Social platforms bury external links and prioritize on-platform engagement, turning corporate accounts into expensive billboards nobody sees.
What still works tells you everything about the constraint. Employee-led social works because faces create trust that logos can't. Influencer partnerships work when someone with an existing audience bets their reputation on your product, transferring trust rather than manufacturing attention. Community-driven growth works because users believe other users in a way they'll never believe your marketing. Every tactic that's working in 2025 is a trust-transfer mechanism, not an attention-capture mechanism. The channels that scaled by interrupting people—display ads, cold email, corporate social—are dying because interruption without prior trust now reads as spam.
And then agents arrived and made the whole thing explicit. When a user delegates a purchasing decision to an AI agent, they're initiating what the research calls a business-to-algorithm transaction. The agent doesn't see your landing page. It doesn't watch your demo video. It doesn't respond to your brand colors or your social proof section or your limited-time offer. It scrapes structured data, evaluates options against the user's stated criteria, and acts. Amazon sued Perplexity on November 4, 2025, charging that its AI agent was accessing Amazon's systems without authorization—evaporating billions in ad revenue by bypassing the display layer entirely. The lawsuit signals the end of an era where companies could compete by optimizing for human attention. Agents represent the ultimate banner blindness.
Inside companies, the trust erosion is just as severe but harder to see. AI tools promise to make teams faster, but the speed comes with a hidden tax: the erosion of accountability structures that used to signal reliability. Code ownership used to mean a specific engineer could tell you why a system behaved a certain way. Now code gets generated by AI, reviewed by AI, deployed by AI, and when something breaks nobody knows who's responsible because the commit history is a wall of agent-generated diffs. Incident response used to mean paging the person who built the thing.
Now it means asking five people who all say "the AI wrote that part, I just approved it." The organizational structures that used to build institutional trust—clear ownership, repeatable processes, humans who could be held accountable—are being replaced by systems that are faster and cheaper but generate no trust because there's no one to trust.
The instrumentation gap is where this becomes a product problem, not just a philosophy problem. According to research analyzing 4.2 million survey responses, exit surveys reportedly get 15.5% response rates while satisfaction surveys get 8.4%. Event-triggered surveys reportedly get 1.6x more responses than URL-based surveys. Single-choice questions reportedly get 3.6x more responses than open-ended ones. The specific moves that work when trust is the constraint are knowable—but almost nobody's set up to measure them. Companies have dashboards tracking activation rate, conversion rate, feature adoption, revenue per user. Almost none of them have instrumentation for trust.
The Lovable Score offers one early model: a composite qualitative metric that tries to capture the gap between "I have to use this" and "I love using this." It's not perfect—no qualitative metric ever is—but it's an attempt to close the loop between shipping features and knowing whether users actually give a shit. The insight is that a lot of "healthy" products are really just products people tolerate because their boss told them to use it, or they haven't found a better alternative yet, or switching feels too annoying. None of that means they love it. And the second a real alternative shows up, they're gone.
Here's the playbook gap. Companies are over-investing in AI tooling—agents that write code, agents that answer support tickets, agents that generate content—while under-investing in trust frameworks. They're optimizing for velocity without instrumenting for the thing that actually predicts whether velocity compounds or just churns faster. They're launching referral programs without tracking whether referred users trust the product enough to convert. They're building AI features without asking whether users trust the company enough to turn on the AI. They're flooding distribution channels without realizing the channels are already flooded and nobody trusts anything they see there.
The industry hasn't converged on how to measure trust at the same velocity it ships features. Net Promoter Score tried, but it's too coarse and too gameable. Satisfaction surveys tried, but satisfaction doesn't predict retention when users are just tolerating your product. Exit surveys work better because negative feedback is always more actionable, but you have to catch people at the moment they're leaving, when the feedback is freshest, and most companies don't instrument that moment at all.
Meanwhile the distribution channels are collapsing in real time. SEO traffic is evaporating as AI-generated answers intercept clicks before users ever reach a website. Paid search is getting more expensive as the total number of searches stops growing. Corporate social gets buried by platform algorithms that punish external links. And now agents are arriving that bypass every channel entirely, making purchasing decisions in a headless commerce flow where brands never get a chance to make their case. Amazon's true fear isn't Perplexity scraping product data. It's that if agents handle purchasing, humans never see Amazon's sponsored listings, and the retail media network that generates billions in high-margin ad revenue disappears overnight.
When creation becomes free, the bottleneck moves entirely to one place: the moment when a stranger decides to trust you with their attention, their money, or their data.
The shift is structural, not cyclical. For most of industrial history, the moat was technology—if you owned the machine, the patent, the manufacturing process, you owned the market. Then the moat became data—the proprietary accumulation of information about users and behavior that took years to build and couldn't be replicated. Then the moat became network effects—your platform was valuable because everyone was already on it, and leaving meant leaving everyone behind. Each of these moats held for a generation. AI collapsed all of them simultaneously by making creation nearly free. According to industry analysis, a functional web application that would have required three months and a team of engineers in 2019 can reportedly be built by one person in a weekend.
A product that required significant development capital several years ago can reportedly be prototyped for a fraction of that cost today.
When creation becomes free, the bottleneck moves. It moves completely, entirely, with brutal simplicity, to one place: the moment when a stranger decides to trust you with their attention, their money, or their data. That moment has not gotten cheaper. It has gotten more expensive, because the supply of things asking for trust has exploded while the human capacity for trust has stayed exactly the same as it always was.
The companies that win the next five years won't be the ones with the best AI features or the cleverest growth hacks. They'll be the ones that figured out how to build, measure, and instrument trust at the same velocity they ship code. That means exit surveys triggered at the cancellation flow. Event-based surveys fired right after a user completes a meaningful action, not when they land on a URL. Single-choice questions that create momentum before asking for the heavy lift of open-ended feedback. Qualitative metrics like the Lovable Score that try to capture whether users actually love the product, not just whether they're using it. And organizational structures that preserve accountability even when AI is doing the work—clear ownership, incident response that can trace decisions back to humans, cultural values that make trust legible instead of treating it as a nice-to-have.
Right now, most companies are building for the Attention Economy while competing in the Trust Economy. They're optimizing landing pages for conversion when agents will never see the landing page. They're launching referral programs that attract low-trust users who churn in two weeks. They're flooding SEO channels that users stopped trusting six months ago. They're shipping AI features without asking whether users trust them enough to delegate meaningful decisions to an algorithm.
The instrumentation gap is the opportunity. Nobody's closed the loop yet. Nobody's built the dashboard that tells you whether you're building trust faster than you're shipping features. Nobody's proven that you can measure trust with the same rigor you measure activation rate. But the early signals are there—exit surveys, event-triggered feedback, qualitative composites, behavioral proxies for whether someone trusts you enough to take the next action. The company that figures it out first won't just win their category. They'll define what product instrumentation looks like for the next decade, because trust is the only moat left that AI can't collapse overnight.