Team I'm on is mandating AT LEAST 50 PERCENT OF ALL NEW CODE be generated using copilot. No longer "use copilot to work faster" but now "replace your development skills with copilot". I do not doubt, FIS will be rolling this out to ALL of development over time
If there were any doubts that FIS was taking the path of using AI to reduce workforce, well. That seems to be EXACTLY where were headed
Can't wait for the brain and skill rot from forced use of copilot to bite them in the a-s
You are all missing the key information. They are in process of bringing in a third party (specialized in AI) to execute AI. They know the data is too scattered across outdated systems and this is too important to leave to FIS staff because they dont trust we have the deep level of expertise required to stand it up successfully. This is critical to the plan. They are paying 10s of millions. You thought McKenzie emptied the wallet ... thats a drop in the bucket compared to what she is willing to pay for this. Her neck is on the line with this.
I have access to a variety of models with GitHub Copilot in FIS. I can select from ChatGPT, Claude, Grok, or Gemini.
It is useful for helping with certain tasks… Generating helper methods, unit tests, SQL scripts, Docker files, Terraform files, fixing dependency issues, parsing errors from large log files, etc.
I still review any of the output manually and I wouldn’t trust it yet with larger features, deep legacy code changes, or architectural decisions. I have seen instances where the developer places too much trust in it though…
As with anything else, the concept of garbage in, garbage out applies. You still have to have knowledge of how the system should work and exactly what you need to include in the prompts or the responses will not be helpful. Thinking you can replace existing engineers for cheaper talent + AI will backfire.
@1fn you can guarantee these mo--ns saw the Block news about replacing workforce with AI and are having big dreams now
We dont allow AI here!
FIS legal and compliance will permit none of those tools or use cases
Recording meetings was too much for them
@en I would second that, Claude code is amazing, don't let the experience of Microsoft co-pilot, all variations put you off, whatever your level, developer or non developer, you ai knowledge will soar, and it's a valuable and marketable skill to learn.
Start with Claude desktop, with Claude cowork, it's only recently dropped
Claude Code and Gemini Code are fully capable of handling one-off Jira tickets and user stories. They require oversight in architectural decisions, but for most tasks presented to developers, they can handle it.
GitHub CoPilot however, is not capable of any of this. Taking a year to get that set up in the enterprise environment (and paying GitHub all the $$$ required to do so) has backed leadership into a corner where they have to make best use of a subpar product. This slow shift also drove away talent who saw the writing on the wall as far as industry migration to AI coding tools goes.
FIS is ultimately too large to ever drive AI development in a major way. But forcing engineers to use a product that ultimately can't perform as well as an average senior engineer is crazy.
@ba The reason you probably think you can only perform simple coding tasks with AI is because FIS only allows Copilot, rather than an AI built specifically for development like Claude. The difference between Copilot and Claude is like hiring a college student who has taken CS101 versus hiring an experienced team consisting of a senior developer, QA automation engineer, DBA, and DevOps specialist.
I’ve seen Claude build an entire website with a database backend for clients to schedule appointments, along with a fully robust administrative backend. It also included scripts to automate the build, testing, and deployment of the application. All of this was completed within a few hours using only a short list of requirements.
If you have a computer at home, I highly recommend installing the free version of Claude and starting to learn. This technology is coming faster than a freight train.
@a4 I have only heard talks of generating code changes from User Stories, as well as a proof of concept done on a very simple change. I have serious doubts that AI will perform to expectations on anything larger than the simplest of changes.
I do not think any of the higher ups understand that LLMs are little more than an overgrown Autocomplete. Microsoft marketing is doing wonders.
Have been singled out for ‘not making regular use of copilot ‘ - embarrassing, how has it come to this
@OP for those who are not developers, can AI code adequately under these circumstances? Having to use their sh---y systems is bad enough. Having to use worse would be diabolical. Familiarity with IVR systems reveals that they work only as well as programmed. End users can find them challenging. Ugh, these cheap SOBs.