AI Tools and Contracts: Picking the Right Model for the Job

Warning: AI and Contracting analogy I haven’t seen yet.

If you’ve been online with others in your field for longer than 5 minutes, you’ve probably seen colleagues arguing over how wrong different AI models are. Some won’t do math for you, none will give you 50,000 words of perfection in one click, some won’t give you humanoids with less than 11 fingers, and some just make up stuff because making up stuff is what “generative” means to a non-techie like me.

But while this prior Contracting Officer is “not technical,” my monthly AI bill is my third largest, right after rent and food and way before air conditioning in a Florida summer.  That’s because, just like when I was a Contracting Officer, I love to experiment and prototype to figure out what works best for different things and how I can get a better result with slightly different words or sequences of words.  I’m always trying new things, and that brings me joy, and I’m always cycling through new models, even if I’m not looking for one model to rule them all.

Today’s social media threads I’ve seen about AI have devolved into name-calling.  Are AI tools perfect?  Nope.  But much of the time—for me, at least—it’s a matter of choosing the right tool or model and figuring out the right prompt to get what I need.  And yes, I still use “Please” in my prompts, not because I fear my future robot overlords, but because I don’t want to forget to be nice to my fellow humans.

My experimentation has shown me how different models work and which ones give me the results I want—and which ones cost too much if I’m scaling a project.  For me, it’s Omni if I want an article summarized or even delivered to me in bullet points. It’s Gemini Pro if I want to pull specific information from a 400-page document I’ve written to find a quick answer and put it into a template for me. It’s Sonnet 3.5 if I need to change every reference in my document from first person to third person or present tense to past tense and make sure it still flows.  It’s OpenAI Playground if I want to transcribe audio files I dictated while walking next to a busy street.  If I want to write about something traumatic, then maybe I’m using Mistral. If I don’t mind the expense, it’s Strawberry if I need “reasoning” to point out potential holes in a complicated document or to suggest, in detail, related topics that I might want to explore and write about from my expertise, based on my resume. If I do mind the expense, then maybe I’m using a free experimental model instead.

All these may change tomorrow or when the next model comes out, but today—this very day—different models are best for different things for me.  What’s best at math?  Ha! I don’t do math, and I waited a lot of years to not have to do math every day or to need to.

My point is, different models are good at different things, and picking the best model is more likely to get you the output you want for the outcomes you need.

Just like picking a contract type.

See?  I told you this was a Contracting analogy, but hopefully a good one for my more technical friends here.

Your Contracting Officer knows which “models” of contracts—whether FAR-based or statutory, like Other Transactions or Procurement for Experimental Purposes—to use to get the best results.   Just because you’ve used Firm Fixed Price contracts a billion times before—like using only Chat 3.5—doesn’t mean it’s the best contract type for what you’re buying.  Maybe this time, you need a different contract type, and a different kind of solicitation than you used in the past, and maybe a few specifically worded paragraphs in the contract to get to the solution you need in the long run or to get to that solution super-fast.