Incumbents often turn new technology into a feature. But a real revolution occurs when technology helps us advance our ways of working. Soon, out-of-the-box AI implementation in products will be table stakes and not a differentiator. It will serve as an assistant for all things you do on your computer. Thoughtful implementation of AI into your workflow and curated/proprietary data and information will be much bigger moats than the ability to train a model.
As the world rapidly adopts new technologies, it’s important to remember that true revolutions don’t happen simply because new technology becomes available. Instead, they occur when societies adopt new behaviors and ways of thinking that fundamentally change the way we live and work. GPT-4 is a prime example of this principle in action. While it is certainly an impressive technological achievement, the inflection point will be when it enables new behaviors and workflows that were previously impossible.
With its ability to generate human-like responses and understand natural language, GPT-4 has the potential to transform a wide range of industries and workflows, from product management and marketing to research and development.
In this article, we will explore the narrow field of AI in productivity tools. Tools that can enable higher productivity in the future of work and allow users to spend more time and effort on what they do best: Nuanced opinion.
The future involves reimagining our most important tools from scratch. For instance:
🗃️ Folders: Why do we need to organize our data like we organized our physical files in cabinets? Is AI the end of organizing?
📽️ Slides: Why did we turn projector slides into their digital equivalents?
🔢 Spreadsheets: Why did we transform mechanical ledgers into their digital equivalent calculators?
When we get new technology, we begin by making it fit our old ways of working. Then, we change our ways of working to fit the technology.
Sometimes, it leads to unintended consequences:
🚙 Ford Model T was invented to help make transportation faster ➡️ But it led to urban migration, the creation of suburbs, and increased traffic
📱Social media was invented to bring people closer ➡️ But it led to people being more distant than ever, mental health issues, and a rise in scams
This happens especially with faster adoption. Speed is the enemy of mindful product build.
Human psychology and technology interact in ways that are hard to predict. In some cases, it can lead to the exact opposite outcome.
The TV was intended to be an information and knowledge multiplier:
“Clay Shirky compares modern sitcoms to gin during the Industrial Revolution. People didn’t know what to do with their lives, so they drank and drank and drank. Now, they watch and watch and watch. Only later did society wake up to new ways of living, made possible by the Industrial Revolution.
Shirky argues that something similar has happened since World War II. But this time, the social lubricant wasn’t liquor but sitcoms. We spent most of our free time watching TV. Now, with the Internet, we have a giant cognitive surplus. Americans watch 200 billion hours of television every year. Meanwhile, the whole Wikipedia project — every page, every edit, every line of code, and every translation — represents the result of roughly 100 million hours of human thought.
Every year, we therefore devote ~2,000 Wikipedia projects to watching television. What if we could transfer some of that energy into something more generative?” ~David Perell’s Friday finds
So, we need to be careful while implementing technology at scale as we can rarely envision the second and third-order effects.
The first step toward thoughtful implementation is starting from first principles (by understanding the user needs and jobs-tobe-done), instead of treating the new technology as an add-on or hybrid with the existing systems.
As we know from one of my previous writeups, a typical user follows this journey in most work-related tasks (can skip a few steps):
And today, most of the AI tech in productivity, especially by Microsoft and Google, is focused on a simple LLM layer on top of existing products like mail, documents, presentations, and sheets. It primarily covers a narrow part of the workflow around creation (see image above; as of early 2023).
Usually, because a more thoughtful approach to implementation would require them to build their product from scratch (which has millions of users with acquired behaviors).
So, incumbents force innovations to be a feature. And when something is built for everyone, rarely does it delight anyone.
🗃️ Big tech turned Dropbox’s online storage into a feature
💻 Instagram turned Snapchat’s stories into a feature
🧩 Microsoft has turned Notion’s block-based system into a feature
Now, Microsoft and Google are doing the same with generative AI. The retrofitting creates two fundamental problems in the creation experience:
- Today, generating the desired output is a trial-and-error process since the creation experience is a black box. AI models take a natural language prompt as input and transform it into text/media/code. But you have little idea of how the output is going to turn out. And it might require multiple retakes to reach the desired output. Good interfaces let users build a conceptual model that can predict how the input affects the output.
- Soon, generating high-quality content will become difficult as expert creators start gating their content. (Peter Nixey, a top 2% contributor on StackOverflow spoke out recently about it)
With this, when a large number of users start using AI mainly for content generation as opposed to other parts of the workflow like research/format conversion it leads to real problems on both the creation and consumption side:
- Opinion problem: When you’re writing a doc/deck/email, you’re sharing your opinion. Asking an AI to write it for you, makes you skip the unbiased thought process and the nuance required to best address the objective at hand.
2. Attention problem: Attention is scarce. The more the volume of docs/decks/long-winded write-ups are created the more difficult it becomes to separate the noise from the signal. In high-value use-cases, AI used irresponsibly, is merely generating junk faster. (Already happening on Reddit)
3. Effort threshold problem: We need to have just the right amount of challenge for users to be in flow and do their best work.
This is the experience fluctuation model.
For many users, current AI capabilities make them feel apathetic, bored, or worried over time by using generative AI in workplace productivity. Their skill is low/medium but their output isn’t high stakes (think high-volume SEO blogs). But there are others whose output is critical and they feel anxiety (since their challenge with the task is high).
The sweet spot is balancing high perceived challenge with high skill. It leads users to a flow state — helping them get the best from AI and their knowledge. (This is what Superhuman does with email!)
They should be able to:
1. Automate a mundane task (e.g., make a human-written blog SEO friendly) ▶️ Feel relaxed
2. Or have a sparring partner along who can challenge or substantiate their thought process ▶️ Feel a flow state
Because great products do less, but better.
Think about the products you love and use most often.
Slack. Notion. Figma. Cron. Arc.
One clear value proposition brought to life with product features, interactions, branding, and communication.
Modern, thoughtful products need to be focused. That means saying no to feature requests and riding the latest tech wave.
We need to understand the parts of the workflow where AI can be the best possible version of itself.
What could you do faster and in a more efficient manner if you had 100s of smart interns with you? 🤔
Because GPT4 is a reasoning engine. It is only as good as the instruction and information we give it.
These are usually parts of the workflow with ‘set-up’ or ‘clean-up’ related tasks that are well-defined, fact-based, or done 100s of times in the past.
Here are some use cases where AI works well: (non-exhaustive)
- Conversion/Summary: Converting a fixed set of knowledge/information from one format to another. Notes to summary. Doc to presentation outline. Presentation to Tweet. Better for condensing information than expanding on it.
2. Breakdown: Lisiting down component parts/next steps of a well-known process
3. Specific search/research: Researching for specific information on the internet (e.g., financial data of Fortune 500 software companies) or searching internal databases (e.g., ‘What was our revenue in Q2 last year?’)
Organizations and individuals who store and catalog their own thinking and data will be at an advantage. They can make those resources available to the model and use it to enhance the intelligence and relevance of its responses.
We are still in the inception era of AI. This era will see the birth of 1000s of new products and intense competition among them since the barrier to entry is low and because AI will drive behavior change (as it enables a 10x bump for users and makes them more receptive to a transition).
In the end, we will have a few, thoughtful horizontal products and many vertical products (with strong moats due to proprietary data and a better understanding of/integration with workflows)
Apple, though relatively quiet at the moment, is well set up to be one of the horizontal leaders.
Which workflows/tools do you think AI can impact most effectively?