I’ve been an avid Cursor user for years now, starting from its early days when it was a simple VS Code-based IDE with simple AI features. The key feature that made me fall in love with Cursor was Tab autocomplete, especially after they relaunched the new model for Tab mode. It was so good. It could predict not only what I wanted to change but also where I wanted to edit next. From that point on, I’ve been building with Cursor.
At work, I was one of the first people asking for paved Cursor support. I was already using it for personal stuff, so I wanted the same productivity at work because moving from Cursor to Copilot was such a big downgrade that I could not take it. Thankfully, Cursor is now one of the paved path tools we use at work.
Even though Cursor has gone through many versions of its UI, and there were definitely days and weeks when I opened Cursor and was confused by yet another new UI change, I stuck with it because it felt like an indispensable tool with a strong emotional attachment to it that I could not explain. The last iteration of the UI, where Cursor launched its Agent Window, feels like a UI that might finally stick. Or will it? I think it fills most of the gaps that had me using the Codex app alongside Cursor.
So I thought I would finally write down some of the features I still miss from Cursor that I would love to see the team build. One note: for some of the ideas here, I have a sense of what I want, but I don’t always know exactly how they should fit into Cursor. So take some of this with a grain of salt.
Activity view
I would love an activity view that shows my most recent running, completed, and past threads, similar to the Codex app. This is probably the thing I miss the most. The current workspace/directory + pinned conversation model feels outdated to me. Once you start running many agents, it becomes almost impossible to organize all your conversations in a way that is not overwhelming while still keeping important work easy to find.
I don't really want to organize conversations. I want Cursor to remember what I have been doing. When I open it, I want to see the living stream of my work: what is running, what just finished, what needs my attention, and what I was thinking about yesterday.
Artifact annotation
I want an easier way to annotate markdown files and other artifacts and pass those notes directly back to the agent for another loop. A lot of my work with agents is no longer just writing code. It is reviewing plans, architecture docs, diagrams, specs, and other artifacts. The loop right now still feels awkward: generate something, read it, describe what is wrong in chat, regenerate it.
I want to highlight a sentence, draw around a section, leave a comment saying "this doesn't make sense," and send the whole thing back into the agent loop. I also want it to be a versioned artifact, so I can see how it evolves over time and, if needed, share its history with my team. This kind of history helps my team understand how I am thinking, why I made certain decisions, and how the artifact changed as my thinking changed.
Living artifacts
Related to that, I would love a way to publish artifacts under my Cursor profile (or a different product that Cursor might as well build) so I can easily share them with other people for comments, changes, and versioning. I know Cursor has Canvases, but that is not quite what I mean. It supports publishing, but to me it still feels like a static artifact that exists at one point in time.
What I want is something closer to a living object. The artifact keeps changing as I think, agents work on it, teammates comment on it, and new versions come out. The conversation, comments, and artifact should all feel connected instead of being separate things spread across different tools. Maybe Cursor Canvas can become this place.
As an example, I wish more people published personal sites with their own writing. Is it really that people don’t have much to say in public? I don’t think so. I think publishing written work on the web still requires too much setup and maintenance. Why does publishing a blog post still feel like real work? Why does keeping a personal site alive require so much effort?
I think Cursor should seriously explore this space, especially as authenticity, personal touch, and taste become more important. We are likely going to have more things written and built on the web than ever before, and I would love for Cursor to make it much easier for people to put their own work out there.
Help me figure out what is worth building
This one is more vague, but probably more interesting. As software becomes cheaper and easier to build, ideas, taste, and deciding what is worth working on become more important. The bottleneck slowly moves toward, "What should I build?”. The number of options is almost endless, which reminds me of a quote I read from Naval, though I’m not sure if he was the original source: "In the age of abundance, we seek scarcity, and in the age of scarcity, we seek abundance." This feels true in almost every field, and I believe it is becoming true for us too, as engineers and builders.
Cursor already sees a huge amount of context about how I work. It sees my prompts, projects, half-finished experiments, half-baked thoughts, questions, code, and the things I keep coming back to. Could Cursor help me notice the ideas that keep appearing in my work? Could it say, "You have explored this idea four times over the last three months. Maybe there is something here worth building?" Could it connect two ideas that I never realized were related?
I strongly believe that not everything in Cursor has to start with an empty agent prompt.
From Coding Agent to Living Workshop
I have found that most of Cursor's recent features have focused on very practical user needs: cloud agents, which I use and love, of course, router, reviews, and generally making software generation faster and easier. All of these are useful, and I am sure Cursor talks to many users and hears a lot about what people want in their workflows. But I also wish Cursor would build more features that create a stronger and more personal connection between the human and the computer.
The first version of Cursor was basically Cursor the IDE. It was built on top of VS Code and felt like a familiar IDE with a new AI layer: Cursor Tab first, then agents inside the IDE. The current version feels more like an Agent Factory. It is a factory whose main job is increasingly to start agents, give them tasks, watch them work, review their output, and spin up more agents. I totally get why the product is moving in this direction, as most agent tools seem to be heading there too.
I think there could be another version after the Agent Factory, which I think of as Cursor as a Living Workshop.
I want opening Cursor to feel like walking into my workshop or studio. You know that feeling when you walk into your office in the morning with your first coffee, ready to do stuff with a lot of energy and excitement? It has unfinished things, strange experiments, tools you have shaped over time, notes from yesterday, and something sitting in the corner that suddenly becomes useful for another project. You move things around, combine them, drop them, come back to them, and once in a while something turns into work you really care about. Often, though, you don't even know what it will eventually become.
My dad is an active abstract painter now, and when he starts a new painting, if I ask him, "What are you painting, Dad?" most of the time he doesn't really know. He is going with the flow. He only knows his next step. He has his paints, his brushes, and maybe some idea of the colors he wants to use, but he doesn't know what the final painting will look like yet. I think there is something important in that.
I feel Cursor can be that studio, where AI is the main material inside it. When I enter Cursor, can I feel surrounded by my ideas? Can I see what I was playing with yesterday? Can Cursor bring back an old prototype because it connects to something I am working on today? Can I throw an idea somewhere without knowing yet if it deserves to become a real project? This connects back to my earlier point about helping spark new ideas.
Ideas are becoming the bottleneck. What can Cursor do to unblock that bottleneck? Right now I feel Cursor is trying to maximize how much software we can generate. Another way to look at the product is to maximize the feeling of growing a living workshop where ideas start, connect, change, and eventually become a few things you really want to build.
Admittedly, this one is hard for me to describe. Not everything that feels intuitive is easy to put into words, but I hope the idea comes through.
Cursor as an AI fluency layer for companies
At the team and company level, I think Cursor can play a huge role in AI fluency. Inside almost every company, there are now a few engineers who become very good at working with AI. They build their own workflows, prompts, rules, MCP servers, skills, scripts, and habits that make them much more effective.
The problem is that most of this knowledge stays with those people. Maybe one engineer has a setup that makes him 1000x productive. Another has built a great debugging workflow. Someone else has found a skill or MCP server that saves hours every week. Other engineers in the same company often have no idea these workflows exist.
Cursor is in a unique place to see these patterns. Can it identify the AI-enhanced engineers and workflows that work unusually well, then help spread those workflows across the team or company? Instead of every engineer having to figure out how to become good at using AI on their own, Cursor could become part of how the company itself learns to work with AI.
Continually improve my harness
I also wish my Cursor kept learning how I work and suggested ways to improve my setup. Based on how I prompt, organize, move between agents, accept, reject, and correct their work, it could suggest ways to make my harness better.
Maybe it notices that I keep explaining the same thing and suggests turning it into a rule or skill. Maybe it sees that one workflow keeps producing bad results and suggests changing it. Maybe it notices that another skill or tool could remove several prompts I write every day.
The product should not only make the agent better. It should make me better at using the agent.
Learn my taste
If taste and judgment become some of the most important engineering skills in the AI age, Cursor should learn mine. I’ve read about how Cursor runs offline and online evals to improve things like CursorBench, so I assume they already have a lot of user signals they can learn from. It also feels technically possible to learn from local Cursor transcripts and the many small choices I make while working.
Today, much of that learning disappears with the thread (AFAIK). I want implicit learned preferences. Over time, my Cursor should understand that I prefer certain kinds of APIs and abstractions, certain testing styles, and certain ways of writing docs. Eventually, two engineers starting with the same prompt should perhaps get slightly different results because Cursor understands their taste.
Cost as a product feature
Organizations are starting to introduce cost limits for employees. We are moving from the first phase of AI adoption, where the goal was simply getting engineers to use these tools, toward a phase where companies care much more about impact and cost.
Is Cursor's automatic model routing enough to meaningfully reduce token usage across a company? Maybe. But I would love to see Cursor go much further here. Can it understand which workflows are wasting tokens? Can it notice when expensive models are being used for tasks where cheaper models usually produce the same accepted result? Can it help companies optimize for cost per accepted change instead of simply tokens used?
I think companies could pay a big premium for this feature alone.
Agent replay and debugging
When an agent screws up, I shouldn't only see its final diff. I want to replay its work almost like a debugger and see what it believed, which files it looked at, what assumptions it made, which tools it called, and where things started to go wrong.
Then I want to change one assumption and re-run the agent from that point instead of starting everything again.
As agents run for longer (which I think is where industry is heading towards) and do more on their own, understanding why they did something becomes almost as important as seeing what they did. Today we mostly debug the output. Eventually, we will need to debug the agent itself. What’s the debugger (we all loved and used in the IDE age) in the age of long-running agents? I don’t know, but I bet it would be super useful to have it.
Final thoughts
Maybe what I am describing is a version of Cursor that becomes less like an AI-native coding agent and more like a place. A place where I work with agents, but also where ideas build up, unfinished work stays alive, and the product slowly learns how I think and what I care about.
The exciting part of Cursor has never been only that it makes writing code faster. It is that every once in a while it gives you a glimpse of a different relationship between humans and computers, one where the computer feels less like a tool you command and more like a space where you think, build, and explore. Smalltalk-80 had a similar kind of impact decades ago: it helped shape many of the ideas behind modern graphical computing and how we interact with software today.
I would love to see Cursor keep pushing further in that direction.