Superhuman's latest auto-draft feature is the first time an AI email assistant has felt less like a gimmick and more like a genuine collaborator. In our testing, the generated replies often required little to no editing, which is a bar that previous iterations have consistently failed to clear. This is not about the thrill of novelty; it is about the quiet utility of a tool that respects your time. For anyone who has spent years refining the delicate art of the email reply, the promise here is not that AI will replace your voice, but that it will handle the cognitive load of the obvious stuff, leaving you to focus on the messages that actually matter. This development feels like a natural counterpoint to the cautionary tales we have explored, such as the experience of talking to my AI clone, which taught me to question the tech, where the novelty of the interaction masked deeper concerns about authenticity.
The practical implication for our readers is straightforward: the threshold for adopting AI in daily workflows has just moved. When a tool requires you to rewrite most of its output, you are simply doing the work twice. But when the output is consistently close to what you would have written yourself, the calculus shifts from curiosity to convenience. This is the moment we start trusting the technology, not because we are asked to believe in its potential, but because it has demonstrated competence in a high-stakes, high-frequency task. We would tell a reader who is skeptical to try it on a low-stakes thread first. See if the tone matches your intent. If it does, you have just reclaimed a few minutes of your day. If it does not, you have lost nothing but a moment of experimentation. The bar is low for a reason: most people are not looking for a revolution, they are looking for a reprieve.
This is also where we see a clear divergence from the broader AI conversation. Much of the discourse is dominated by abstract fears or inflated promises. The Forrester function, often discussed as a mathematical curiosity, has found relevance in machine learning, but that kind of technical abstraction rarely translates into tangible user benefits. Superhuman's feature is the opposite: it is grounded entirely in a single, measurable outcome. Did you have to edit the draft? If the answer is no, the feature works. That simplicity is the real innovation. It is not about the sophistication of the model or the elegance of the interface; it is about the reduction of friction. And as we have seen with tools that verify AI’s understanding during tax season, the most valuable applications are often the ones that solve a specific, annoying problem without adding new ones.
The open question we are left with is not whether AI can write a good email, but whether we will let it. The feature works, but it will only succeed if it becomes a habit. The risk is not that the drafts will be bad; it is that we will stop paying attention to the words we send out under our names. That is the detail to watch. As the tool gets better, will our standards for what we are willing to send without a read-through drop? If they do, we have not saved time; we have just outsourced our judgment. The best outcome here is not a future where you click send without thinking, but one where you spend your energy on the messages that require a human touch, knowing the rest are handled. That is the promise worth exploring.