This as told essay is based on a conversation with Brandon Pedersen, a 41-year-old senior principal engineer at GoTo, based in the Salt Lake City area. The following has been edited for length and clarity.
As a senior principal engineer at GoTo, where I started working in 2013, I am the technical lead of an AI enablement team responsible for maintaining AI platforms, creating new tools that help people do their jobs, and providing training and consulting support across the company.
On any given day, I might be helping train a team to write better AI prompts or automate parts of their workflow through an AI skill or custom application. We also collect use cases from people across the company who want to automate tasks. We evaluate what will have the biggest impact on the business and decide what is worth automating and how we can help teams build their own applications.
I also look at the architecture and systems involved in what we build in other teams. Because I’ve been at GoTo for a while, I can work across departments, like marketing and sales, to connect the pieces.
Here are some of my tips for deciding when to use AIbased on my experiences.
1. Don’t blindly trust AI
From the beginning, whether it was generating a single line of code or a simple function, I always reviewed AI’s output as there were a few errors here and there. I think those errors have decreased in the smaller use cases, but they still show up in larger ones.
Compared to three years ago when I started working heavily on AI, I trust that our AI tools, including Claude Code and GitHub Copilot, can get meaningful work done for me, allowing me to build large-scale solutions, create larger presentations and web pages, and help with full applications. My confidence in AI’s ability to handle larger tasks has improved tremendously.
2. Be precise with your directions
When I look at an AI output, I assume there is probably something in it that could be improved, clarified, or corrected. That doesn’t mean it isn’t useful; it means that I approach it with a critical attitude.
I’ve always viewed AI through the lens of an experienced engineer, spending a lot of time reviewing code and pull requests and noticing changes that others might miss.
I have found this same attention to detail useful when evaluating the results of AI. I saw what these tools delivered, what mistakes they could make, and how small changes in the wording of a prompt can completely change the output. Once I understood that, I became more conscious about when to use AI, how to use it, and how much trust to place in its responses.
3. Do the mission-critical work yourself
I was tasked with building our internal AI governance policy. At first I thought we could use one of the publicly available industry standards, ask AI to collect the information and generate a solid draft policy for us, but I chose not to. I knew that this policy would be shared internally, that there would be critical reviewers of the document, and that I would have to explain and defend every word of it.
When you ask AI to generate such documents, it can produce something that looks pretty good at first glance, but once you start analyzing it more closely you realize that it doesn’t always make sense in the day-to-day reality of how your business works.
In this case, I decided to write the policy myself and use AI to help with things like diagrams and supporting artifacts that made the concepts easier for people to understand, rather than the actual policy language.
4. Know when to delegate – use the two-second rule
There are certain things I see people trying to automate that would take them two seconds to do themselves. I wouldn’t use AI for personal communication with others. We don’t want to become a proxy for an AI model. If someone asks you a question, they can ask AI themselves, but they are looking for your input and don’t want to feel like they are talking to a machine.
I also wouldn’t use it to make judgments about people – not just because of legal risks, but because AI doesn’t share our values and doesn’t have the full context.
You don’t have to automate everything in your work. There are many things you could automate with AI, but you end up wasting time and resources fighting with the tool, waiting for it to respond and going through the process when you could have just done it yourself faster.
5. Validate and verify factual claims
I also learned that sometimes you have to push back on AI rather than accept the final answer. One of the best examples of this came from a training that our finance group shared more broadly throughout the organization. The advice was simple: if AI gives you an answer, ask about the sources of the information, follow the links and check the logic.
I recommend doing this especially for questions that worry you, especially unfamiliar topics, because you don’t know what you don’t know. Even just having the thought to distrust the outcome, not to assume it is smarter than you, and not to be blinded by the impressiveness of the response is good because then you won’t miss it if it leads you astray.
Even if you’re not entirely sure it’s wrong, it helps to challenge it and say, “I don’t think that’s right, show me your sources.” Sometimes it will catch itself and correct its own answer. If you want to use AI more consciously from now on, this is a very practical first step.
6. Start experimenting today
As soon as people actually start using AI, they get excited. In 30 or 40 minutes they can sometimes solve a problem they are stuck on, just by describing in simple language what they want to do. This gives people the feeling that they can make changes themselves, preserve what they have built and then share those ideas with others.
As we look to the future, AI will advance in capabilities, speed and intelligence, and will likely make fewer mistakes, but it will remain our job to use our own judgement, criticism and values to get the best out of it.
