Ten years ago, I started systematically tracking emerging technologies as a venture analyst. What began as a job requirement became an obsession, then a career. Looking back at a decade of tech scouting, some patterns and lessons stand out.

Here’s what I’ve learned about spotting the future.

Lesson 1: The Future Is Already Here, Just Not Evenly Distributed

William Gibson’s famous quote has become cliche, but it’s deeply true. Almost every technology that will matter in five years exists today, somewhere, in early form.

The skill isn’t identifying things that don’t exist. It’s recognizing which existing things will scale.

When I started tracking AI seriously in 2016, the core technology of deep learning already worked. The question wasn’t whether neural networks could do useful things - they already could. The question was whether they’d become practical for widespread use. Understanding that distinction would have saved me from both missing opportunities (being too skeptical about AI progress) and bad bets (being too optimistic about quantum computing timelines).

Practical implication: Don’t look for things that don’t exist yet. Look for things that exist in labs, research papers, or niche applications that could expand. The signals are there if you know where to look.

Lesson 2: Timelines Are Almost Always Wrong

I’ve made hundreds of predictions about when technologies would reach various milestones. I’ve been right about the direction far more often than about the timing.

Technologies consistently take longer to mature than optimists expect. Then they spread faster than skeptics expect once they do mature.

The hype cycle captures this - technologies are overestimated in the short term and underestimated in the long term. But knowing this doesn’t make timing predictions easier.

Practical implication: Build strategies that are robust to timing uncertainty. If your bet only works if technology X arrives by exactly 2027, you’re probably making a mistake. If your bet works anytime in the next decade, you have more room for error.

Lesson 3: Technology Is the Easy Part

Most of my timeline errors have been because I focused too much on technology and not enough on everything else - business models, regulation, infrastructure, organizational change, user acceptance.

Self-driving cars exist technologically. They don’t exist practically because of regulations, infrastructure, liability, and edge cases. The technology was ready years before deployment.

AI agents exist technologically. They’re deploying slowly because organizations need to build processes, training, oversight mechanisms, and trust.

Practical implication: When evaluating emerging technology, spend as much time on non-technical factors as technical ones. Ask: “What else has to change for this technology to be adopted?” The answer is usually more than you think.

Lesson 4: Most Predictions Are Extrapolations

When I review my old predictions, the ones that were wrong usually failed because I was extrapolating from current trends without accounting for discontinuities.

Extrapolation works until it doesn’t. Trends reverse. S-curves level off. Black swans appear.

COVID accelerated certain technology trends (remote work, digital commerce) and decelerated others (travel tech, physical retail innovation). No amount of trend analysis in 2019 would have predicted this.

Practical implication: Extrapolation is a starting point, not an answer. Always ask what would make your extrapolation break. Have scenarios for discontinuities.

Lesson 5: The Experts Are Often Wrong About Their Own Field

Domain experts have deep knowledge that’s valuable. They also have blind spots, vested interests, and anchoring to existing paradigms.

Some of my best predictions came from noticing when expert consensus seemed to be ignoring contrary evidence. Some of my worst came from deferring too much to expert opinion.

Practical implication: Consult experts, but weigh their input against their incentives and track record. Independent thinking matters even (especially) in technical domains.

Lesson 6: Follow the Talent

Where smart, ambitious people choose to work is one of the strongest signals about where value will be created.

In 2014-2016, watching top computer science PhDs choose to work on deep learning rather than traditional software told me something. The talent flow anticipated the AI boom.

Today, I track where top technical talent is going. It’s not a perfect predictor, but it’s a useful one.

Practical implication: Pay attention to hiring patterns, PhD topics, where YC companies cluster, what ambitious young people are excited about. Talent flows signal opportunity.

Lesson 7: Skepticism and Enthusiasm Aren’t Opposites

The best tech scouts I know are simultaneously skeptical and enthusiastic. They can be excited about long-term potential while clear-eyed about near-term limitations.

Too much skepticism makes you miss real trends. Too much enthusiasm makes you fall for hype. The balance is an ongoing calibration.

Practical implication: Practice holding multiple views simultaneously. “This technology is overhyped right now AND will be transformative in ten years” are compatible positions.

Lesson 8: The Second-Order Effects Matter Most

The most important implications of new technologies are usually not the obvious ones.

Smartphones were obviously useful for communication. Less obviously, they created an always-connected population that made ridesharing, mobile payments, and social media possible.

The interesting questions aren’t “what will AI do?” but “what will become possible because AI exists?”

Practical implication: For any emerging technology, spend more time thinking about what it enables than what it does directly.

Lesson 9: Conviction Should Vary

Some things I have high conviction about; others I don’t. This isn’t wishy-washy - it’s calibrated uncertainty.

I have high conviction that AI capabilities will continue improving. I have low conviction about which companies will capture the value. I have high conviction that climate change will drive technology investment. I have low conviction about whether direct air capture will ever work economically.

Treating all predictions as equally certain is a mistake.

Practical implication: Explicitly state your conviction level alongside your predictions. Act differently based on conviction - bet big on high-conviction views, hedge on low-conviction ones.

Lesson 10: The Work Never Ends

A decade in, I’m more uncertain about many things than when I started. The more I learn, the more I realize how much I don’t know.

This isn’t discouraging. It’s humbling in a useful way. The posture of a learner - curious, open, willing to update - serves tech scouting better than the posture of an expert.

Practical implication: Stay curious. Keep learning. Update your views when evidence warrants. The moment you think you’ve figured it out, you’ve probably stopped seeing clearly.

A Closing Thought

Tech scouting isn’t about being right. It’s about being useful - providing perspectives that help people and organizations make better decisions in conditions of uncertainty.

After ten years, I’ve been wrong many times. But I’ve also helped people avoid bad bets, seize good opportunities, and think more clearly about the future.

That’s the job. The future will keep being uncertain. The work of trying to understand it continues.

Here’s to the next ten years.