ChatGPT

Open ChatGPT, ask it a question, and you'll get an answer.

That's why ChatGPT blew up. You don't need to know prompt engineering or model names. You just need to know how to have a conversation. But the reason the chat experience actually sticks is because it's memory. It remembers who you are, what you care about, and what you have been working on, so every new chat does not feel like starting from zero.

I use it to pull ideas together and create drafts, but I also feed it things I've written, so it can do a final pass for clarity. It's worth flagging that ChatGPT can be too agreeable if you don't push it. Ask for blind-spot feedback. Tell it when you don't want a rewrite. Make it challenge the piece instead of politely smoothing everything out.

ChatGPT is the only AI tool I reliably use for voice + video chat. I use it all the time for random real-life problems. My washing machine breaks, and I can ask, "What is this error message?" while ChatGPT's camera is pointed at the machine.

Random chats have a time and place, but Projects are where ChatGPT starts to feel more like a workspace than a bunch of disconnected conversations.

We use projects for writing, product, engineering, where the same context needs to show up again and again. If you have a recurring type of work, for us, that looks like writing software reviews in a specific voice, it makes sense to keep the relevant context, instructions, and conversations together. It's also nice that it keeps a specific POV, so you don't have to feed it context and re-explain things every single time you need something.

I do wish we didn’t have to go to the step of linking conversations to projects though. The same goes for skills. I want to access the skill without adding it to a specific project first. It feels like a small workflow thing, but it would cut so much friction if the right context showed up more naturally.

So, I like Projects. I use them. I would recommend teams use them. But the workspace layer still feels like it can use a lot of improvement. It's useful and heading in the right direction, but it still asks you to organize more manually than I would like.

It's kind of shocking to say this because cross-device support feels like something every AI tool should get right, yet I've run into way more friction with other tools than with ChatGPT. Across desktop, browser, and mobile, ChatGPT has easily been the most reliable experience I've used.

Right now, I'm a huge fan of the browser, ChatGPT Atlas. I tried the Claude extension, but lasted two days because f you accidentally close it out for a second, it deletes your entire chat and there is no memory. Feels like a huge oversight and unless you experience ChatGPT Atlas, you don't realize how easy it could be to have AI always accessible on every page.

ChatGPT can connect tools like Slack and Drive, then help you find information or take action across them. The idea is to use AI to pull the answers together instead of having to search across five different tools.

The most useful thing I've found in Agents is honestly just Slack search. I hate Slack search, so being able to ask ChatGPT "Who said this in Slack?" or "Where is this message in Slack?" without adding in context is genuinely useful.

But a lot of agent stuff still feels early. Sometimes it pulls up things where I am like, "I don't even know what this is." And that's the issue with so many AI agents right now. They can connect to your tools, but connection doesn't equal context.

We test sooooo many tools trying to do this agent thing, so maybe I'm just harder to impress here, but it doesn't feel new or exciting to me. Maybe for someone who's seeing this for the first time, it will, but for me, I can't help but notice the gaps. For these types of things, I am still more bullish on using specific AI tools that are more refined.

If you mainly want an agent-style assistant, I'd look at something more specific, like Littlebird. It has workflows that are better because it's actually looking at my screen instead of relying solely on connected tools.

I asked our Head of Engineering about his views on Codex.

He said "It's the best value in agentic coding right now. The $20 plan gives you enough usage to get things done, which you genuinely can't say about Claude's equivalent tier."

He reaches for Codex when he needs something fast and obvious done across a lot of files, the kind of work where it's straightforward and hard to mess up.

He also runs it alongside Claude specifically to audit each other's work.

Because they're somewhat adversarial, they're good at spotting what the other one got wrong. If your budget is tight and you need one tool for real development work, he says to start with Codex.

Something I love so much about ChatGPT is that it knows how to handle a mess. I can bring it transcripts, PDFs, spreadsheets, and all my half-baked ideas. It'll take all that raw material and clean it up way faster than I could.

I can't even count the number of times that I've brought ChatGPT, a rambling transcript, and a bunch of research notes, asking if it could turn it into a useful review! It's good at pulling out structure, themes, contradictions, and usable lines when I'm too close to the material to see what is actually there.

However, as much as I am a fan, I still wouldn't treat it as a final source of truth. It can miss context, overstate things, and sometimes it struggles with being too polished, so you still need to pass over everything with a fine-tooth comb and some editorial judgment. But it still saves a lot of time, especially when you've been staring at a wall of text and need to know, "What's actually here?"