Agentic Data & Software Engineering

The pace of AI tools is frantic: your team doesn’t have to chase it

By Xavier Franco, CEO de Elternativa · Grupo Zemsania
· 5 min read

Written with the help of AI and reviewed by Xavier Franco

A walkway of flat stones crossing a fast-flowing river, with white water breaking around each stone.
In this article

If you lead a technology team, this will sound familiar: every week a new model appears, a new AI tool, or a version that changes everything. And every time there’s someone who wants to try it, someone who has already installed it, and a third person asking whether you should switch. I say it plainly: the pace of AI is frantic, and your team doesn’t have to chase it.

The pace isn’t going to slow down

It helps to start by accepting reality. Stanford’s 2025 AI Index gives a figure that sums it up: the cost of using a model at GPT-3.5 level fell from 20 dollars per million tokens in November 2022 to 0.07 dollars in October 2024. More than 280 times cheaper in about 18 months.

When something gets cheaper at that rate, new products keep appearing. And adoption is already massive: according to the 2025 DORA report, 90% of technology professionals use AI in their work.

So the question isn’t whether there will be another novelty next week. There will. The question is how much of the way you work changes every time one arrives.

In practice: count how many different AI tools your development team uses today. If the number surprises you, keep reading.

The hidden cost of chasing every novelty

Trying a new tool seems free. It isn’t. Every switch drags along:

  • Learning again how to ask for things, what it does well and what it does badly.
  • Rewriting the instructions, conventions and shortcuts the team already had.
  • Losing what was learned, because it lived in the previous tool’s configuration or in the head of whoever used it.
  • Mixing ways of working, with some people on one tool and others on another, and the same problem solved two different ways.

Multiply that by every novelty and every person. That’s the real cost of chasing the pace.

What changes and what has to stand still

My proposal is to separate two things we tend to mix up:

  • The tool changes. Claude Code, Cursor, Codex, Windsurf or whatever comes out tomorrow. It makes sense that it changes, because they improve every month.
  • The method stays. How a task is defined, the criteria for accepting it, who tests it, who validates it and where what was learned ends up.

If the method lives inside the tool, every tool change is a project. If the method lives in a system of your own, changing tools is almost trivial: the new one receives the same rules and carries on where the previous one left off.

In practice: ask where your team’s conventions for working with AI are written down. If the answer is “in the tool’s configuration file”, they’re in the wrong place.

A role for each tool

What works isn’t choosing the best tool, but giving each one its own role. Some are better at defining and talking with the business; others at coding in parallel and unattended; others at reviewing. With a common method, each tool knows what its job is: analysing, building, testing or validating.

That way a novelty stops being a threat. If a better tool for coding comes out, it takes that role with the same rules as the previous one, and the rest of the system doesn’t notice.

Use the AI you already pay for

There’s one more effect I think matters. Many companies already have corporate AI under contract: Claude, ChatGPT, Copilot, Gemini. Chasing every novelty usually ends in more licences, not more results.

With a common method, you first get the most out of what you already pay for. And when a new tool really adds something, you bring it in knowing which role it takes and which data you’ll compare it with.

In practice: before approving a new AI tool, ask for the role it will cover and the metric it will be compared on against the current one.

The method lives in the system

In our case, the rules are written once, at their level: what comes built into the platform, what’s common to your company, what belongs to each client or department and what belongs to each project. What you define at the top applies to everything below. And at every step, the system gives each instance the rules that apply to it, without depending on what it remembers.

That’s why it works the same with one instance as with twenty, and with today’s tool as with tomorrow’s.

Where to start

  1. Take inventory of the AI tools your team uses.
  2. Write down outside the tools the conventions that live inside them today.
  3. Give each tool a clear role.
  4. Compare before switching, with metrics from your repository.
  5. Let the method decide, not the novelty of the week.

How we help

At Agentic Data & Software Engineering we orchestrate your AI development platforms so they work in parallel, unattended and governed, without your team changing tools: Claude Code, Cursor, Codex, Windsurf or VS Code to develop, and Claude Cowork, ChatGPT Work or Gemini Enterprise to manage.

We do it with Agentagile, our methodology for developing with AI. We don’t build agents: you use the AI you already have under contract and Datagentis turns it into a team that works with method. The tool changes; the method stays.

Frequently asked questions

Do you have to switch AI tools every time a better one comes out?

Not necessarily. If the method lives outside the tool, you can bring the new one in for the role it covers best and compare it with data before switching.

Which AI development tools can be orchestrated?

The ones you already have under contract, such as Claude Code, Cursor, Codex or Windsurf, each with its own role within a common method.

Where should your AI working conventions be kept?

In a system of your own, organised by levels (company, department, project), and not in each tool’s configuration.

How much has the cost of AI fallen?

According to Stanford’s 2025 AI Index, using a model at GPT-3.5 level went from 20 dollars to 0.07 dollars per million tokens between November 2022 and October 2024.

Keep reading