If you feel like every vendor, headline, and LinkedIn post is talking about AI, you are not imagining it. That overwhelm has a name, and it is AI fatigue. Getting past AI fatigue is more doable than it looks. At a recent BVS Biotech Connections evening, hosted at Beacon CoLab in the Boston area, Bay State IT’s Sean Sanker Jr. and Jude Stiel of baobab.bio gave a talk on that exact feeling. They spent about half an hour walking through what is actually useful day to day, what is mostly marketing, and a few habits anyone can try the same night.
You can watch the full talk below, and we have pulled out the main takeaways underneath.
Why AI Fatigue Happens
Part of the problem is volume. There is simply too much information about what these systems are and what they can do. Terms like “agent” get thrown around until they lose meaning, and every company seems to be selling one. We wrote more about that feeling in our earlier post on overcoming AI fatigue.
The other part is frustration with the tools themselves. People run into two recurring issues:
- Non-determinism. Ask the same thing twice and you can get two different answers. Run a familiar task on a new dataset and the results wobble.
- Hallucinations. The model states something false with full confidence.
As Jude explained, this is not a bug you can fully remove. A language model is a statistical model of language, and some randomness is built in. You would not want it to flip a coin and land on heads every single time. The real question is how to take something unpredictable and get reliable work out of it. That is where a handful of simple habits help.
Tip 1: Use the Best Model and Turn On Thinking
One of the easiest wins: switch on thinking mode. Sean and Jude recommend everyone use it. In ChatGPT, click the model dropdown in the upper left and actually choose your model. Skip the instant option and use a thinking or pro model. In Claude, try Opus instead of Sonnet, and make sure the thinking toggle is on.
Why does this matter? Jude ran a quick demo with the audience. He asked people to answer instantly, no thinking allowed: “What do cows drink?” A good chunk of the room said “milk” even though the answer is water. A model forced to respond instantly makes the same kind of snap-judgment error. Giving it room to think lets it talk itself toward a better answer before it commits.
Larger models help too. They are more sample efficient and simply more capable on a given task. The upgrade costs you nothing but a couple of clicks.
Tip 2: You Are the Expert, So Be Specific
A common myth is that you can type “make me the best slide deck in the universe” and get exactly that. You will not. The model does not know who you are, so it optimizes for an average that fits no one in particular.
The fix is specificity. These models are generalists. They hold a broad compression of human knowledge, but they are not specialists in your field. You are. When you supply your background, your constraints, and the details only you know, you pull far more value out of the tool. You can also leverage this while creating skills or custom GPTs, so the agent can continually reference your knowledge.
Tip 3: Ask the Model to Interview You
When you give the model a task, add a line like: “List every assumption you are about to make and every default you will use, then ask me any blocking questions before you start.”
In Claude this is especially smooth, because it can present the questions interactively. Answering them narrows the field from a generic, average result down to something built for your situation. It is one of the fastest ways to turn a weak first draft into a strong one.
Tip 4: Give Real Feedback, and Save Your Work
If you get a bad result, you are not stuck. Telling the model “you did it wrong, do it better” rarely helps, just like it would not help a colleague. Instead, explain specifically what went wrong and ask what it needs to do better. Does it need more context? Better source material? A different tool?
A few more habits that pay off:
- Save summaries. When a session goes well, ask the model to save a summary as a text file. Upload it to future sessions so you are not starting from scratch.
- Start fresh for new topics. Do not ask for dishwasher instructions in the same chat you used for two weeks of unrelated work. Open a new conversation for a new task.
- Know that memory is not learning. When a tool “remembers” details about you, it is not changing what it is. It is just attaching notes to the conversation. You still have to supply the right context each time.
A Word of Caution Before You Install Anything
Here is the part that matters most for any organization. Tools like Claude Code are powerful precisely because they run on your machine and can read files, execute code, and connect to the web. That same power is a real security consideration.
Please do not install Claude Code or similar agent tools on a work device without talking to your IT team first. An agent that can read a folder can also read a file full of Social Security numbers it was never meant to see, or delete something permanently. In a work setting, these tools need to be reviewed and properly secured before they touch company data. The same caution applies to any AI service: know where your data is going, and stick to reputable platforms. If you are not paying for a product, you are often the product.
This is exactly the kind of rollout where our AI deployment team helps companies, from choosing the right tools to setting policy and training staff so the tools get used safely.
Wrapping Up
Getting past AI fatigue does not mean mastering every new product or chasing every announcement. It means understanding enough to feel steady instead of overwhelmed, picking a couple of habits like turning on thinking and asking the model to interview you, and remembering that the plain chat window already delivers real value long before you ever touch an agent. Experiment, stay curious, and verify what matters. You will not break anything, and you will learn a little more each time.
If your team is feeling the same AI fatigue and wants help building a practical, secure AI strategy, reach out to our team. We provide guided training as well as other AI solutions and would be happy to help you find the next right step.
