SoftwareSeni has been selectively evaluating tools across the AI landscape since ChatGPT was announced in late 2022. Our developers had experimented with LLMs even before that (gpt-3, the predecessor to GPT3.5 that was the original “ChatGPT”, was opened to unrestricted access in November 2021).
The tools themselves have started to improve rapidly over the last few months as newer State of The Art (SOTA) models and “reasoning” models have been released by all the major providers – OpenAI (makers of ChatGPT), Anthropic (Claude), Google (Gemini).
The SOTA models provided by these companies power almost every third party AI tool, particularly coding tools. And the effectiveness of those tools has grown along with the power of the models.
At the same time, both the model providers and the tool companies have matured to the point where they now have secure offerings.
The model providers have been using interactions with their models (either via their user interfaces like ChatGPT.com, Claude.ai and Gemini.google.com, or programmatically via their APIs) to gather more data to train their models on.
If you ever wondered what the little thumbs-up and thumbs-down icons under the responses in ChatGPT were for – that is it. Volunteer model training.
No business wants their IP or their customer data copied, used in training an AI, and possibly appearing in later outputs of that AI as a part of a response to some random person’s query.
Important industries could never be clients to these model providers or use any tool built on them. For example, finance and healthcare have stringent data handling requirements.
But there’s a lot of money in those industries, so now the model providers offer their paid services with privacy guarantees, including fine-grained permissions management at the enterprise level.
Tool providers, like Cursor, who make a code editor, have also made privacy part of their product and provide detailed explanations on how they interact with your code. This is part of the reason they are valued at $10 Billion, despite being a fork of the open source Visual Studio Code editor (created by Microsoft) and their reliance on third party SOTA models for generating code.
The AI tools SoftwareSeni uses
With our focus on software development we obviously have a huge interest in coding tools. Our teams work across a wide range of platforms, languages and frameworks, so we weren’t able to pick just a single solution.
We ended up going with Cursor and Github Copilot. They’re currently in competition to achieve feature-parity with each other, with Cursor being a little cleverer and a little more agentic at the moment.
But Copilot has the advantage of running across multiple IDEs, meaning it is integrated into the tools our developers already use, like JetBrains IDE. Cursor is just Cursor – you use it in as part of their VS Code fork or not at all.
Both tools provide strong privacy guarantees in their paid versions.
Away from coding, in general operations (recruitment, training, etc) we will be using ChatGPT. We also considered Claude, but OpenAI’s range of models and team features, as well as a bit of first-out-the-gate advantage, has made them our first choice.
There are differences between the platforms, but like the coding tools, everyone is racing towards feature parity. And with AI to write code for them, any gaps between competitors shouldn’t last long, right?
Choose your own tools
Our clients, particularly our extended software development team clients, can choose the tools they want developers to use. Or they can choose to have their developers continue to code traditionally if they aren’t comfortable with the privacy guarantees from the model and tool providers.
We do have active clients that have already moved their software development teams over to AI coding tools.
Despite what you might hear or read about people rapidly building apps, “vibe-coding” their way to new products in days on their own, reliable, resilient, scalable software still requires professional developers and time. If, as they say, the devil is in the details, then software remains pretty much angel-free. Though testing and documentation does get quite a boost – two things developers love to neglect but make every project better.
Here is a pair of tweets that demonstrate the upside and downside of using AI coding tools without the necessary experience:
Is your team ready for AI?
The next few years are going to be very interesting. Interesting for the whole world, but especially interesting for everyone like us who is building a business around software. We’re going to be able to achieve more and move faster. Small teams are going to be able to take on projects that used to be out of reach.
If you add AI coding tools to your team, we expect a few months in, if not sooner, you will start re-assessing your timelines and your product goals.
Get in contact with us if you want to talk about an AI-powered software development team or project. We’d be happy to chat through the options and possibilities with you.
