At lunch today, I caught most of a conversation from the table next to me.
They were probably in the same industry—this area is full of tech companies. The first half was all complaints about the changes AI was bringing, saying software is becoming less and less valuable, that they now need two or three times as many clients to hit the same performance targets as before, and that they feel more exhausted.
Then the conversation suddenly shifted. Someone suggested building their own replacement for some HR-related service. Another person immediately chimed in, saying they knew someone with resources, and wondered if they could turn it into a business.
They had been anxious just moments before; now they were getting excited.
Anxiety and Excitement—We All Have Both
It’s not just that table. In elevators, at meals—pretty much every conversation you overhear these days is about AI. My friends are the same: anxious that their old rhythm has been disrupted, excited that there might be new opportunities.
These two emotions alternate for everyone. I think this is just the new normal.
Nine Days Without a Release and People Start Asking Questions
The pace of change really is accelerating.
Today OpenClaw dropped its latest release. Before that, there had been nine days without an update, and lots of people in the community were asking: What’s going on? Are they cooking up something big?
Nine days. People felt that nine days of silence was worth asking about.
They did drop something big in the end—permissions and the ecosystem were completely rebuilt. But I’m more interested in the rhythm itself. Back when we made software, planning in weeks felt normal. Now, if an idea hasn’t shown any sign of life after a week, people basically stop paying attention.
Bottleneck #1: Passion
At this speed, what actually blocks startup teams?
I think the first thing is passion. Not the motivational-poster kind.
Things are changing too fast, and attention shifts just as quickly. Trying to keep going in this wave purely for profit is unrealistic. Machine production speed has gone up, but demand hasn’t kept pace. People still make decisions slowly; consumers are too lazy to think or switch. This inertia is actually the biggest source of stickiness right now, while the technology itself offers very little moat.
You have to genuinely enjoy the work itself. Otherwise you won’t last. I later wrote a piece specifically about this—AI Doesn't Amplify Skill, It Amplifies Passion.
Bottleneck #2: Focus
Twenty percent of energy here, thirty percent there, forty percent in yet another direction. That used to be just barely manageable. Not anymore.
The entire industry feels like a hundred-meter dash. If the core team is still running at a marathon—or even a race-walking—pace, they’ll quickly find people passing them in waves until they disappear from sight.
Direction follows the founding team’s level of commitment. Whether a small team can fully dedicate itself to one thing—this focus directly determines whether they can keep up. Teams with divided attention won’t survive long at this speed.
Bottleneck #3: Infrastructure and Process
Some bottlenecks are surprisingly not technical; they’re in very basic places.
The product is ready, you’re about to launch, and then you discover the domain needs ICP filing—5 to 20 business days. Overseas? The server is ready the moment the domain is registered. That single step is an order of magnitude slower.
The same goes for internal approval processes, the speed at which you can spend money, and the decision-making chain. These used to be just a little slower, and it didn’t matter much. But when technology is moving at this magnitude, everything around it that can’t keep up becomes an obvious obstacle.
Bottleneck #4: Physical Stamina
This is one many people don’t expect.
People assume AI should make everything easier. But that’s not the case right now. AI still can’t run fully autonomously 24/7 without human oversight, so the drain on people has only grown.
You can spin up five agents working in parallel, building and testing at the same time. But human attention has to switch rapidly between multiple things, constantly checking results and course-correcting.
Honestly, the mental overhead of managing a team of agents is greater than managing a team of real people. It’s like running a high-intensity workshop all day long—not the occasional kind, but every single day. I experienced this during my 300K Lines of Code in 10 Days stretch—after five or six hours, I felt completely drained.
If your body can’t take it, you’ll fall behind quickly.
A Transitional State
My feeling right now is that this is a transitional period.
Once some core directions and value chains start getting validated, you can build a 24/7 AI loop around them so things no longer rely entirely on human energy. Add to that the ability to attract more contributors to co-build, and it’s no longer just one person or a small team pushing forward with tokens—more people’s brainpower and resources can join in.
This is already happening in our own open-source projects. Someone submitted a feature built entirely with an AI coding agent, and it got merged—and this person isn’t a programmer. As long as you have ideas, you can participate; your identity shifts from consumer to builder.
The people at the next table during lunch—anxious in the first half, excited in the second—will probably stay in these two states for a long time. The key is knowing exactly where you’re stuck along the way.