🧙🏼 The AI products I actually used in 2025

--

Author

Dario Chincha
December 21, 2025

The AI products I actually used in 2025

I went from mostly chatbots to mostly command line. Here’s how my AI usage changed this year.

Was this email forwarded to you? Sign up here.

My most-used AI products this year were all from the big labs.

This year, I went from chatbots as my main tool for daily productivity to command line tools (although, I still use chatbots actively, just not for coding).

I definitely caught FOMO a few times this year, bouncing between chatbots, coding tools, and image generators as new and better models were released.

Here’s the AI products I used the most this year, how I was using them at the start of the year and how I’m using them now.

OpenAI’s products (ChatGPT, Codex, the API)

At the start of 2025:

At the beginning of the year, ChatGPT was my #1 tool for pretty much everything except writing. Like most of us, Chat was the first GenAI tool I started using, and I'm still using it.

How the tools changed:

There were a lot of important product launches and updates from OpenAI this year:

Other notable upgrades:

There were a large number of other features and updates in ChatGPT apart from these, none of which I actually use myself.

How my usage changed:

GPT-5.2 is still my go-to model for analysis, research and daily tasks. I use the model for coding (inside Codex) too; it did most of the heavy lifting when I vibe coded my own outbound sales tool. However, I now leave the most complex tasks to Claude Opus 4.5.

I use ChatGPT Projects actively, a feature which was launched in December last year and got upgraded with project-only memory this year as well as shared Projects for teams.

I still use ChatGPT for all kinds of research, but I no longer use Deep Research. If I want to go deep on a topic I choose GPT-5.2 with Extended thinking enabled; I've found the value to be almost the same.

I was using the image gen in GPT-4o until Google’s Nano Banana was released. However, I haven’t had a chance to try the brand new ChatGPT images yet, but it looks amazing.

What I spent:

I bounced back and forth between Plus and Pro plans for ChatGPT. I also used my ChatGPT plan to power my Codex usage. I used the API (particularly GPT-5 mini) to enable AI functionality in a couple of internal tools that I made. Estimate I spent around ≈$1K on OAI this year.

IN PARTNERSHIP WITH ZENCODER

Most teams struggle to ship fast in a world of shifting requirements, large codebases, and unpredictable AI tools.

Zenflow brings structure and multi-agent orchestration to modern engineering. It gives you a spec-driven workflow powered by multi-agent orchestration, built-in verification, and parallel execution.

  • Your specs become executable plans

  • Agents collaborate on coding, testing and review

  • Every output is validated before it hits your repo

Teams use it to eliminate rework, avoid regressions, and ship features faster.

Anthropic’s products (Claude, Claude Code, MCPs)

At the start of 2025:

Claude started the year as my trusty writing partner. It used to feel like the most creative of the available AI chatbots not very reliable or accurate; that changed over the course of the year.

How the tools changed:

The most transforming upgrades and launches:

  • Claude Code in research preview (February)

  • MCP (April)

  • Opus 4.5 (November)

Other notable upgrades:

How my usage changed:

I’m now using the Claude chatbot, Claude Code and MCPs very actively.

I still use the Claude chatbot as my go-to writing partner, although it has gotten some light competition from Gemini 3 Pro lately. I also use their writing styles actively to write in my tone of voice.

For coding, I bounced back and forth between Claude Code and Codex as Anthropic and OAI kept 1-upping each other in terms of model performance. I’m now back to Claude Code after the release of Opus 4.5; it’s my go-to model for complex coding tasks.

Claude Code was the biggest release from Anthropic this year. Putting an agent inside the terminal completely changes what you can achieve with the underlying technology. It also sparked the launch of OpenAI's Codex and tons of AI app builders, including the unicorn Lovable.

I was a skeptic at first, but have also become an avid user of MCPs; they’re brilliant for connecting coding tools to adjacent tools, and have become much more useful as the industry (including other AI labs) have started embracing the standard. The MCP that most impacted my work this year was the Supabase MCP; it allows Claude Code and Codex to access, read and write to my database.

As an example of the power of MCPs. I’ve told you earlier about how I’ve vibe coded my own sales tool. This week I asked Claude to create a ranking of 1,000+ sponsor leads and sort my prospects based on how well they fit my newsletter. In one prompt, it used the MCP to read my DB’s records, analyse them, and update the records with their ranking.

What I spent:

I switched several times between Claude Pro and Max plans. I use my Claude subscription to power my Claude Code usage. I’m back on Max as of this week (need more of that sweet Opus power). I did some experiments with their API but went with OpenAI in the end for my current workflows. Estimate I spent ≈$500 on Anthropic this year.

Google’s products (Gemini, Nano Banana, the API)

At the start of 2025:

I wasn't using Gemini at all. In fact, I had a strong aversion to ever touching any AI made by Google after the intensely sub-optimal experience that was Bard. That firmly changed this year.

How the tools changed:

The most transforming upgrades and launches:

Other notable upgrades:

How my usage changed:

I successfully abstained from using Gemini until Nano Banana came along. That's when I admitted to myself that Google is actually making useful things again, and it has since become the model I use for image gen.

I was also impressed by the Gemini 3 Pro model, especially for writing; I'm using it alongside Claude now for a second opinion.

I was using Antigravity with Gemini 3 Pro for a couple of weeks after it came out and found it to be really good. However, once I tried Claude Opus 4.5 and GPT-5.2 inside Codex, it was over for Gemini coding-wise for me. That might well change with their next model release though.

The brand new Gemini 3 Flash is in the API is awesome. I’ve just changed from GPT-5 mini to this model to power certain AI functionality in my own apps and saw big speed improvements + cost reductions on some input-heavy tasks (with same level of output quality).

What I spent:

≈$20, from using the Gemini in the API.

For my Gemini app usage, I’ve been riding on the free plan. If I was using image gen a lot, I would have upgraded to be able to access Nano Banana Pro, but my use is very occasional.

Google has been on a roll with all its releases lately and I expect to be using Gemini (especially the API) a whole lot more in the coming year.

❦

The biggest game-changer for me personally in AI this year wasn't more powerful LLMs per se, but moving into Command Line Interface (CLI) tools. If you’re still only using web interfaces, I really recommend trying Claude Code, Codex, or Cursor. There’s a learning curve, but vibe coding with direct access to the command line is the most rewarding hobby project you can take on right now.

🛠️ My vibe coding guide: I keep a live list of the most popular vibe coding tools (and what I’m currently using). It’s unsponsored and mostly just for my own tracking, but feel free to use it to help you get an overview and choose a tool to start with. Check it out.

Next year, my goal is to keep bringing you AI news with zero hype and showing you how I’m using it, but also expanding more into how this community and companies are getting results.

With that, I hope you have great holidays and time to properly recharge. Thank you for letting me be part of your inbox!

See you in 2026,

Dario

IN PARTNERSHIP WITH THESYS

C1 by Thesys turns any n8n workflow into a smart, adaptive AI app - with interactive UIs instead of walls of text. From chatbots to AI agents for research, analytics or automation, no coding and no changes to your workflow logic. Thesys is the UI your n8n workflows have been missing.

THAT’S ALL FOR THIS YEAR!

Was this email forwarded to you? Sign up here.

Want to get in front of 21,000+ AI builders and enthusiasts? Work with me.

This newsletter is written & shipped by Dario Chincha.

Disclosure: To cover the cost of my email software and the time I spend writing this newsletter, I sometimes work with sponsors and may earn a commission if you buy something through a link in here. If you choose to click, subscribe, or buy through any of them, THANK YOU – it will make it possible for me to continue to do this.