We are living in the most disorienting moment in the history of technology discourse, and tech is an industry that has always veered between both. But this is next level. At any given moment, your social feed will tell you that artificial intelligence is about to end civilization, that nobody will ever work again, that we are 12 to 18 months from a permanent job apocalypse …

… and then, next scroll, that large language models are nothing more than pattern matchers, talking parrots, probability machines regurgitating stolen content, incapable of original thought. The pendulum doesn’t just swing between hype and doom. It whips back and forth so fast that even people who work in this space (myself included) can find it genuinely difficult to keep their footing.

The problem isn’t that any one of these narratives is entirely wrong. It’s that they’re all partially right, stripped of nuance, and amplified to maximum volume by people who have strong incentives to occupy the extremes. The AI doomers need existential risk to justify their platforms. The hype merchants need transformative promise to justify their valuations. The dismissers need AI to be trivial to justify their decision not to engage. And all of them are louder than the people doing the less loud, careful work of actually figuring out what these systems can and cannot do.

So how do you stay grounded? How do you make decisions, for your business, your career, your team, when the information environment itself is this volatile? Here are some practical tools that have helped me, and that I think can help anyone willing to invest a little effort in their own sense-making.

Test It Yourself. Seriously

The single most grounding thing you can do is use these tools directly, with intention and structure. Not casually. Not once. Consistently, across different tasks, with an eye toward what works, what breaks, and where the edges are.

Ask the same question of multiple models. Give them the same complex task and compare the outputs. Push them until they fail, because they will, and pay attention to how they fail. Do they hallucinate confidently? Do they lose the thread of a long conversation? Do they handle ambiguity well or collapse into generic answers? Your own direct experience with these systems will teach you more than any op-ed or keynote, because you’ll see the gap between marketing claims and functional reality in real time.

You don’t need a research lab to do this. You need curiosity, a bit of patience, and a willingness to document what you find. Keep notes. Build your own mental model of what AI is actually good at in your specific domain, because that answer will be different from the generic narrative.

Diversify Your Information Diet

If everyone you read agrees with each other, you’re not informed, you’re captured. The AI discourse is dominated by a handful of very loud voices, and most of them are selling something: a product, a worldview, a consulting practice, a reason to panic.

And here’s the part that makes this actively dangerous: the platforms themselves are working against you. Social media algorithms are self-reinforcing by design. The moment you click on, linger over, or interact with a piece of AI doom content, the algorithm registers that as a signal and serves you more of the same. Engage with a hype post, and suddenly your entire feed is wall-to-wall AI utopia. You don’t have to subscribe to a single perspective — the algorithm will build an echo chamber of one around you without your conscious participation. This is especially true on X, where the recommendation engine has become aggressive about surfacing content that matches your recent engagement patterns regardless of whether you followed anyone expressing those views. The result is that your perception of the “consensus” on AI can be almost entirely manufactured by what you happened to tap on last Tuesday.

Diversifying your information diet is more than a nice idea. It’s a deliberate act of resistance against systems that are economically incentivized to lock you into a single emotional frequency.Deliberately seek out perspectives that make you uncomfortable. If you lean toward the doom side, read the people building practical applications and shipping real products. If you lean toward hype, read the critics; not the dismissive ones, but the serious researchers who are publishing on limitations, alignment challenges, and evaluation failures. If you think AI is just statistical plagiarism, spend time with the cognitive scientists and philosophers who are wrestling with what these systems actually represent, because that question is far less settled than either side wants you to believe.

Some starting points: follow researchers who publish their work, not just thought leaders who post threads. Read the actual papers when you can, even just the abstracts and conclusions. Pay attention to who changes their mind publicly, that’s a strong signal of intellectual honesty. Be skeptical of anyone who has more invested in being right than being informative, has been confidently predicting the same thing for three years, and insists on never updating their position.

Watch the Incentives

Every piece of AI commentary comes attached to someone’s business model or career trajectory. This doesn’t automatically make it wrong, but it should always make you curious. When a venture capitalist tells you AI will transform every industry within two years, ask what their portfolio looks like. When a tenured professor tells you LLMs are fundamentally incapable of reasoning, ask what their research program depends on being true. When a tech CEO says their model has achieved a breakthrough, ask what their next funding round looks like.

This isn’t cynicism. It’s basic media literacy applied to a domain where the stakes are high and the incentives are enormous. The most trustworthy voices in this space tend to be the ones who acknowledge uncertainty, who distinguish between what they’ve tested and what they’re speculating about, and who don’t need AI to be any particular thing in order for their work to matter.

Separate Timelines from Capabilities

One of the most common errors in AI discourse is collapsing timelines. People take a real capability (say, AI systems that can write functional code or summarize complex documents) and project it forward to a conclusion that may or may not follow: therefore, all knowledge work will be automated within 18 months.

The jump from “this works in a demo” to “this replaces a job” is enormous, and it’s filled with integration challenges, organizational friction, regulatory questions, reliability requirements, and the messy reality of how work actually gets done in complex systems. AI is transforming workflows right now. That is real and worth taking seriously. But the speed at which that transformation reaches any particular job, in any particular industry, at any particular level of reliability, is a much harder question than the loudest voices want to admit.

When someone makes a timeline claim, ask: based on what? What would have to be true for that to happen? What are the dependencies? What’s the track record of similar predictions? You’ll find that most confident timelines are built on assumptions that haven’t been tested.

Embrace the Uncomfortable Middle

The hardest place to stand in the AI discourse is the middle, where things are genuinely impressive and genuinely limited, where the technology is both transformative and overhyped, where legitimate concerns about job displacement coexist with legitimate excitement about new capabilities. The middle doesn’t generate clicks. It doesn’t get you invited on panels. It doesn’t reduce neatly to a headline.

But it’s where the truth lives. And it’s where the best decisions get made, without panic, without euphoria, but with a clear-eyed assessment of what’s in front of you right now, tested against your own experience, informed by diverse perspectives, and held with the appropriate amount of uncertainty.

The AI hype-doom perma-cycle will keep spinning. We all have moments of doom and hype. Your job (unless you’re an analyst or a journalist) isn’t to stop it. Your job is to stop letting it spin you.