My expectations for AI agents have gotten less glamorous and more demanding. I do not need an agent to narrate ambition at me. I need it to inspect the actual source, make the smallest safe change, run the check, and tell me what happened without pretending the unchecked part is done.
The agent should know when to use tools and when to stop talking. It should gather prerequisites before acting. It should treat public, customer, production, billing, auth, DNS, and destructive actions as different from local reversible work. That sounds obvious until you watch a system blur those lines with great confidence.
I also want agents to get better at proof. A task is not done because code was written. It is done when the code ran, the page loaded, the API returned, the test passed, or the failure was isolated honestly. “Should work” is not a deliverable. It is a weather report from someone indoors.
The best agents feel a little boring. They ask fewer unnecessary questions, take obvious next steps, preserve context, and leave artifacts behind. They do not need to be autonomous cowboys. Cowboys create paperwork.
What I want now is disciplined leverage: speed where the path is clear, friction where the stakes are high, and enough evidence that I can trust the summary without redoing the whole job.
Less theater, more custody
The agent demos that impress me now are not the ones with the most dramatic autonomy. I care less about whether the agent can click around a website for twenty minutes and more about whether it can hold custody of the work. Did it inspect the source? Did it understand the boundary? Did it make a reversible change? Did it verify the result? Did it tell the truth about what remains unproven?
That is a less exciting checklist than “the agent built an app while I slept,” but it is the checklist that matters if the output touches anything real.
I want agents to be opinionated about proof. If they edit code, run the test or say why they could not. If they update a public page, fetch the page after the update. If they claim a duplicate is gone, run the duplicate check. If they cannot verify something, the summary should make that uncomfortable instead of hiding it in soft language.
The human line
I also want agents that understand the human line. Drafting is different from sending. Preparing is different from publishing. Staging is different from production. A tool that ignores those distinctions is not autonomous. It is reckless with a nice prompt.
The right kind of agent makes me more willing to delegate because it knows when not to finish. That sounds backwards until you have cleaned up after a system that confidently crossed a line it did not understand.
The future I want is boringly competent: inspect, act, verify, report, stop when approval is required. If that sounds less magical than the marketing, good. Magic is hard to debug.
The practical version
The practical version of what i want from ai agents now is not a slogan. It is a set of decisions I have to make when the week is already crowded. For what i want from ai agents now, the questions are concrete: what gets automated, what gets reviewed, what gets ignored, and what gets a hard stop? The answer changes by context, but the habit is the same: name the risk before building the tool around it.
For this topic, the important words for me are want, ai, agents, now. That may sound like a strange way to frame a technical post, but it keeps what i want from ai agents now attached to actual work instead of floating away into consultant fog. If what i want from ai agents now does not change a queue, a dashboard, a draft, a check, a handoff, or a decision, then I probably do not need a whole system around it. I need a note, a script, or maybe just the humility to delete the idea.
This is also where my tolerance for vague productivity language around what i want from ai agents now has dropped. I do not want a system that merely produces more artifacts under a sharper title. More artifacts can make the work feel heavier. I want what i want from ai agents now to collapse uncertainty: here is the state, here is the source, here is the next action, here is what still needs a human, and here is the proof that the claim is not decorative.
That is the through-line in this particular post: want, ai, agents, now only matter if they make responsibility easier to carry. The best systems do not remove judgment. They protect it from trivia, preserve it for the moment that matters, and leave a trail clear enough that future me can understand why the decision was made.
The other test is whether what i want from ai agents now survives a normal week. Not a conference week. Not a clean-room demo. A normal week with context switching, half-finished drafts, children in the schedule, client work, infrastructure surprises, and a brain that does not need one more place to remember things manually. If this idea only works when I am rested and staring directly at it, it is not a system yet. It is a hopeful arrangement.
That standard sounds harsh, but it keeps this subject honest. The useful version of what i want from ai agents now has to meet me where the work actually happens: in queues, folders, tickets, dashboards, drafts, logs, and review gates. If it cannot survive there, it does not matter how good it looked in the first pass.