For as long as humans have come up with tools: axes, abacuses, calculators, they have always multiplied the effort and skill of their users. In the hands of a master, they do wonders; in the hands of the unprepared, they cut fingers off. With computers nothing changed: still a force amplifier, a “bicycle for the mind”. Amplification reached new, sometimes transformative levels, but there was never any magic. If what you're multiplying is zero or negative, don’t expect a positive outcome. The rule is solid: garbage in, garbage out.

But in 2025 everyone started talking about agentic agent agency, as if we’d already crossed the threshold from amplifiers to independent creators. AI companies promise a future where technology acts on its own, and many take these promises at face value, rarely asking who those promises are actually really for. Now, users find themselves believing that the future is already here, that AI has truly achieved agency and can read your mind, and create and decide on its own. Some leaders sense potential and rush to invest, hoping to gain an edge if the technology ever catches up with the hype. This only adds more fuel to the bubble and raises the stakes even higher.

AI might reach that point, but we’re not there yet. Meanwhile, the pace and the pressure keeps accelerating. Before LLMs, people built confusing, broken, and ugly websites. Now, thanks to AI, we have 100x more of them, built in record time!

At this moment, AI is still just a tool, an incredibly advanced and truly impressive one, like having Wi-Fi on an airplane. It’s made many things easier: I can now program in languages I barely know, it can quickly process huge amounts of text and predict answers with surprising accuracy. But it’s still a tool. We need to stay in the loop, direct the outcomes, and take responsibility. Agency hasn’t arrived.

Related links:

• What can agents actually do? (by Will Larson)
• Coding with LLMs in the summer of 2025 (by antirez)
• I’m Tired of Talking About AI (by Paddy Carver)