Tell us about yourself. What are you working on right now?

I'm Justin Delisle, cofounder and CEO of Tato. Originally I was a software engineer, then I became a consultant, a solution architect, a technical architect, and ultimately a technical director overseeing operations for a consulting firm in Montreal, in the enterprise resource planning (ERP) environment—digitizing businesses to make them more efficient. 

We were exclusively a Microsoft partner, and I worked with a lot of Microsoft technology before I grew frustrated with large ERP projects and the wider sector—it’s very unsuccessful [at addressing] problems that technology can solve. So, I decided to quit my job and start Tato to solve large enterprise projects, to stop them failing and make them more successful by using generative AI.

How do you use AI for work?

Large ERP projects often fail because of bad stakeholder communication—people forget to follow-up on things, track an action item, surface a risk, or do what a project needs in order to be successful. We see Tato as a project to build a platform that's going to change the world, but we also apply it to ourselves. If this was a Microsoft Teams call, Tato would join and take notes on any risks or decisions, action items, and next steps that come up, so we can be sure that we don't drop any balls. 

It’s been fun using our own technology on ourselves. Of course, it’s a little different when you're smaller; there's only five of us, so we don't have the same challenges as a 50-person team. Still, a lot of the same core principles—in terms of trying to keep up with things—are there, and I'm always excited to try out the latest and greatest [technology]. When ChatGPT came out I purchased a license, and at least 10 times a day I copy and paste something or discuss topics with it, to figure out how I can use generative AI to make me more efficient as a person.

I really care about what we're solving, and I believe generative AI is well-suited to solve a lot of real world problems

How do you use ChatGPT within your team?

When we're talking about anything, our natural reflex is to put it in ChatGPT, ask a question, and then read the reply. We copy, link, and share with each other, and most of the time I want to dig deeper on a question or specify something to see how it influences the question. 

It'd be great to interact with it together, but we both have to interact on our own side and then share the result, which is inefficient—the collaborative experience of ChatGPT is horrible, and I look forward to better collaboration experiences in future LLMs, including having them participate in discussions rather than be a software interaction. 

One of the things I find really annoying about LLMs is all the context you have to provide and the fact that it doesn't learn from you. That's something we're trying to fix at Tato—we want Tato to learn about your context. 

How did you first realize that GPT could help with your workflow?

We don’t have a lot of money, so we cherish what we do have and try not to spend it. Lawyers are extremely expensive, and I have to review a lot of documentation that often includes jargon I don’t understand. Putting it into ChatGPT and identifying risks is helpful—I still pay my lawyer to validate it at the end, but skip them having to point out obvious mistakes in our work. It’s been one of our most common use cases.

When building a startup, you also need to be very concise and clear about what your company does when describing your value proposition. I’ve been brainstorming for that alongside ChatGPT because it’s great at giving different ideas and possibilities for describing ourselves, even if I’ve also found that it’s horrible at being concise. As a startup founder trying to define a business, it’s really nice—you can make it play around with words 40 different times and see what sticks and what doesn’t, then refine your prompts to avoid certain terminology or patterns, or ask it to summarize what it understood. 

It’s a good filter—if it doesn’t understand what I’m doing then I know I’m using the wrong words.

Do you always start by asking it to give you alternatives for how to say something?

Frequently, but oftentimes there might be a word that I know could mean different things in different contexts, so I’ll ask for some potential interpretations. 

I also have ChatGPT play the role of a customer—for example, I might be going to present my platform to somebody who’s in the banking sector, and ChatGPT will give me a lot of good alternatives so I can know, “Okay, they’re in banking, so this is most likely what they’re thinking.” 

It’s helping me understand my users a little bit better. There’s nothing more valuable than speaking to the customer directly, but sometimes it’s 11 p.m. and you can’t, so ChatGPT does the job.

How would you describe the value that using GPT has had for your team?

It explains concepts fast. We have customer calls all the time where they mention something, and ChatGPT allows us to learn about it quicker than we could have before. We would have had to surf the web and find the right source, but there’s no way to do that during a call with a customer. 

Like this morning, a customer said, “Everybody uses EOS.” I’d never heard of it, so I ChatGPT’d it in the background and it told me that “EOS” is a framework that people use to run businesses and meetings, and they have “rocks” [defined goals] that they manage inside their business—and just as I was reading that in the corner of my eye the customer said, “We have rocks that we manage in our business.” 

It’s a lot faster for learning, because I don’t need everything—I just needed that general understanding.

What adjustments do you make most frequently to improve your prompts?

I’m an impatient user. Most of my interactions with ChatGPT have been, at maximum, three [exchanges] back and forth. I think that the quality of the content gets worse and worse as you expand a conversation with ChatGPT, and I feel like it actually understands what I’m trying to get to less and less.

That said, there are a few scenarios where I go deep on the interaction and I have 10 or 15 [exchanges], but that’s one percent of the time, and in those cases it’s decent because they’re things that are very common on the internet, or related to a specific industry—like the legal paperwork, for example. We’re hiring somebody in Bosnia, but I forgot to mention that and just asked about a new employee and their contract, and in ChatGPT’s response it said, “It depends on the location where they’re working.” I refined [the prompt] and said, “The person is working in Bosnia, how does that change what we’re doing?”

If it points out a risk that I don’t understand, I can ask about it; if I forget some details that are important to the interaction, it can remind me to be more specific. 

What’s your top tip for refining prompts?

Setting expectations for what you want is a game changer in LLMs. If you ask ChatGPT to summarize something and you’re expecting to get it in a list format, you have to say, “Give it to me in ‘Header. Bullet point. Bullet point.’” That’s been the biggest change from how I use it today versus how I used to use it in the past—being clear about my expectations because otherwise I’m often not happy with the result.

Do you have a hot take on generative AI?

I don’t believe that micro-interactions solve much—I believe that, for LLMs to help, they have to understand the bigger picture and help you solve those bigger picture items. 

Inside enterprise projects there are meetings, and when most people think about software and technology they want to make meetings more efficient and valuable. But a meeting isn’t just a meeting, is it? A meeting is just a mechanism to get to your overall goal. 

What you need to do is look at how the meeting affects the continuum of information. If something is able to help you with the continuum of information, that’s where you have a breakthrough. Today’s software doesn’t help with that very much. That’s our focus at Tato, and why we’re different.

What aspect of generative AI are you most excited about?

The power of generative AI is that it can see the world through your lens. Humans are very different, and we all evaluate the bigger picture from different lenses. The CFO might want their business to be profitable and standardized and to have better compliance, whereas for the business analyst they want to make the system stable. Generative AI is flexible enough to see a project through you and actually make sense of them in the larger context too.