The impending effects of all these new AI tools
Get ready for some serious changes in who does what at work
You may not be aware of the cambrian explosion of artificial intelligence (AI) models and tools that is happening. As of today, the onslaught of new models and tools built on top of them haven’t quite “crossed the chasm” from the early adopter community of the technophiles to the broader mainstreet population. So while a lot is getting written about what these models do and what changes they are going to cause, it’s largely written by a bunch of people who already accept the future reality these tools will enable.
What is lacking is a clear and concise understanding of how these tools will affect those of us who are not yet using them. So let me take a stab at it.
A great starting point is Tomas Pueyo’s Uncharted Territories article about generative AI, which is a type of AI that can create new data based on existing data. The article discusses the potential applications of generative AI and how people can use it to create new products and services. You’re probably better off reading that than reading any more of my article, if I’m being honest. He provides a treasure trove of rabbit holes you can go down into.
Source: State of AI Deck, 2022.
What will AI do to those of us in the knowledge economy? It will reduce the cost of creating to near zero. The friction involved in making things will be non-existent.
See that dank meme? I didn’t have to do what I normally do when making a meme, which is scroll through my mental rolodex of meme formats, trying to remember which one is the best fit for what I want to express. Instead, I used SuperMeme’s Text-to-Meme functionality by typing in “using AI to make a meme about using AI to make a meme”. The amount of effort it typically takes me to make a meme decreases. Less friction, which leads to easier generation and increased output. While not quite at zero, the cost reduced.
Now, perhaps you’re thinking to yourself “Jesus, what a waste of some smart engineer’s time, harnessing new AI models just to build a meme generator that lets shitty newsletter authors like Cy make their dumb little images a little faster” and you would be right, it is a poor use of AI.
Instead, think about your most loved/hated Excel file for a moment. You know, the one that dictates how your day/week will be at work. What if instead of trying to think through the combination of formulas necessary to give you the answers you wanted (was it SUMIFS followed by IFNA or the other way around?) you just typed in "tell me how many enterprise accounts didn’t log in last month?” and you got the answer in the excel file?
You could ask follow up questions like “What about for specific enterprise accounts?” or “How does that number compare to other months?” or “What were the reasons given for not logging in?”. You can imagine how much faster and easier your work would be if the computer could do the grunt work of data retrieval and analysis for you. This already exists, with tools like Excel Formula Bot. Within the year, it will be de facto norm for Excel and GSheets.
Don’t believe me? Consider that an AI model (GPT-3) wrote the previous paragraph, using the paragraph before the last as its prompt. That’s right, it’s AI writing about using AI. All I did was hit “CMD + Enter” and the word processor I’m using to write this newsletter (Lex) wrote it for me.
AI makes it so much easier for people to optimize their work and improve output. Because AI decreases the friction in building and creating, it will reduce the segmentation that occurs between the folks who naturally gravitate to new tech and those that don’t.
Consider the previous example of the Excel file. In most companies, there is a small population of employees who are more skilled at things like data analysis. They aren’t intimidated by the formulas or error messages you have to fight through to get to the answer. These are the “power users”, whether its Excel, Salesforce, Tableau, SAP, Looker or whatever other software dominates their sector of the company. The power users have the upper hand and the additional job security because they know the inner workings. They know how to whisper the secret incantations to the machine spirits within the glowing box, unlocking the terrifying powers of the VLOOKUP.
For many, there’s too many rules, too much arcane knowledge to acquire. AI abstracts this all away. No more formulas, no more remembering the difference in analytics languages between one software and another. Instead, you just write out plain text like “how many enterprise accounts logged in last month?” and poof! Your answer is there.
Power Users are no longer the gatekeepers. Every employee is a capable analyst. How would that change the competitive landscape at work? Even more so, how would it change how fast a company could move? How would it change decision making when everyone has access to data that was once the domain of the power user?
The impact of this on the jobs market is huge. We’re seeing it already with the rise of low-code and no-code platforms that allow people with no coding experience to create entire websites and apps without writing a single line of code. These platforms will only get better and more powerful as AI gets better at generating code. And as AI gets better at generating code, the more capable it will be of automating the tasks currently performed by software engineers and other highly skilled knowledge workers.
Even more impressively, AI overcomes a lot of the “dirtiness” of data that is systemic in most companies, overcoming errors and omissions. The act of normalizing, cleansing, mapping, or whatever synonyms people use to describe making data sets play nicely with each other will become a thing of the past as models will be able to overcome deficiencies using synthetic data for omissions and automatically correcting errors. This work, often the domain of analysts and sometimes data engineers and scientists, will shift to models.
Armed with new abilities to rapidly translate data into actionable insights, employees will be able to ask and answer new, broader questions that were once too much effort for the traditional, manual processes. The speed at which an organization can adapt and move will increase as everyone can quickly generate new insights.
Organizations will have to learn how to adapt and be comfortable with the new levels of transparency that come from a wider dissemination of data.
Watch your back, Data Analysts!
Designing
With the power of “text to image” models, it will be extraordinarily easy for a layman to build out their own mockups. Will it be good? Depends on what you need it to do. But if it’s enough to align a group on what a thing should look like, it will be worth its weight in gold. Good designers are tough to find (and expensive!). And if we’re honest, far more things get designed than get built. That’s a good thing, as building mockups are an important tool in any lean product development process. With the ability to easily build your own mockups, that process only accelerates. What Figma did to disrupt Photoshop, AI will do to Figma.
Regionalization
Need to add Cantonese to your website to help improve sales in southern China? Need to produce a ton of training material on top to help your professional services team scale? AI models like PaperCup will enable this in a fraction of the time. Perhaps companies won’t need quite as large of a sales org because much of the busy work around localizing the company’s content and making it consumable to target populations is done automatically.
Summary
Imagine how this might work together. For example, say you’ve got a large data set, something like a quarter’s worth of customer behavior on a ecommerce website. We’re talking millions of data points, as every little click, time on page, IP address, and more gets recorded.
AI models will be able to rapidly extract insights, something that would take business analysts and PMs many hours and only after having enough domain knowledge to know what to look for. The long process of hunting for correlation between the populations of users and their many variables boils down to hours if not minutes.
With extracted insights (i.e. customers keep items in their cart an average of 32 hours before making a purchase > $1000), a narrative gets written by another AI model that summarizes all the key findings and how they relate together, forming the basis for future investments in the website. It also converts the narrative format of the summary into bite-sized bullet points and puts it into a slide format for easy distribution to broader set of stakeholders who are impacted by the investment but aren’t directly involved.
A third model takes that summary, finds the key descriptive components, and generates mocks of the next version of the website’s cart page. A fourth writes out the backlog for engineering. A fifth co-writes the code. A sixth translates the mocks to all the other languages the website supports.
Without the weight of manual processes holding them down, software development teams can rapidly iterate and improve faster than ever before. And as a result, companies will be able to move faster and be more responsive than ever. Smaller teams shipping more product faster and at higher quality. Roles that were once entirely about execution moving more towards inspection. I expect there’s a lot of jobs that are about to get shaken up.
Who to Read?
Want to learn more? Here’s a few recommendations
Ben’s Bites - Ben sends a daily email on advances in AI. Great for keeping your finger on the pulse of this industry.
the team at Every - A lot of great authors, including personal favorites Evan Armstrong and Nathan Basche, are covering how AI is making waves. They also write about many other things of interest, it’s a good investment to subscribe. Plus, they’re the creators of the word processor I’m using, Lex!
Alexander Wang just dropped some thoughts on how AI is going to impact warfare. Probably not most readers’ cup of tea, but certainly more important than “AI for dank memes”. Recommended reading so you understand it’s far more than just corporate knowledge work that is getting disrupted by AI. We’re ~10 years away from Skynet.
Generative AI will soon be slipping into what teaches our children; shaping even how future researchers conceive of the field they will enter and the future AI techniques they will create. https://kinnu.xyz/blog/research/the-generative-ai-revolution/?utm_source=substack&utm_medium=email