Whether or not AI will replace humans depends on who you ask. Opinions vary, but part of the difficulty of answering the question has to do with the difficulty of quantifying what even counts as AI from a technology standpoint.
In an interview on the Radio Davos podcast from the World Economic Forum, Head of Artificial Intelligence and Machine Learning at the World Economic Forum’s Centre for the Fourth Industrial Revolution Stuart Russell explained at length how “it’s actually surprisingly difficult to draw a hard and fast line and say, this piece of software is AI and that piece of software isn’t AI.”
According to the World Economic Forum’s Future of Jobs Report, 2020, the time spent on work-related tasks by humans and machines is projected to be equal by 2025. Still, though, it’s essential to evaluate this in context. In a 2023 article in the Harvard Business Review, Karim R. Lakhani, Harvard Business School Professor, echoes a popular sentiment on the matter among business leaders across industries: “AI is not going to replace humans, but humans with AI are going to replace humans without AI.”
These changes are being acutely felt in the software development space, where two approaches are beginning to dominate the landscape: One based on the assumption that certain development workflows will be fundamentally autonomous, and another that is convinced humans will have a continuing and active role in software development, augmented by AI support.
Founded in 2020, Pieces is an AI-enhanced productivity tool that helps developers become more efficient by providing personalized workflow assistance. In a recent episode of the ‘AI in Business’ podcast, Emerj CEO and Head of Research Daniel Faggella recently sat down with Pieces Technical Co-founder and CEO Tsavo Knott to discuss how automation will impact the role of developers, and why Pieces is adopting a more human-focused approach.
This article examines two critical insights from their conversation:
- Adjusting to technological advances: Fundamentals for building a software development career in an industry landscape where automation and augmentation are commonplace.
- Recognizing the limitations of AI systems: Using copilot technologies to complement and upskill to a 10x developer level, and tackling problems that are still difficult for AI, such as technical debt.
Listen to the full episode below:
Guest: Tsavo Knott, Co-founder and CEO of Pieces
Expertise: Coding, Software Development, Entrepreneurship, Interactive Media, Computer Science
Brief Recognition: Tsavo graduated from Miami University in 2018 with Bachelor’s degrees in Game and Interactive Media Design as well as Computer Science. Before co-founding Pieces in 2020, he was a vice president and co-founder of Accent.ai, a language learning platform.
Adjusting to Technological Advances
Tsavo begins the conversation by acknowledging that companies approach automation and augmentation in two distinct groups. To make these categories specific, Tsavo explains that, first, some companies are focused on figuring out if it is possible to automate and replace the roles of certain developers fully. These companies are evaluating whether or not autonomous agents are powerful enough to replace what a developer does outright. Tsavo describes that the approach is being applied to workflows whether the developer is a full stack developer building an end-to-end application or a developer working on continuous integration (CI) and continuous delivery (CD) in DevOps.
He identifies his own company, Pieces, as belonging to the second category that understands that the role of the developer is changing but not to the point where developers are becoming extinct. As CEO, he notes how Pieces recognizes the shift in the role of developers, which means they are becoming more cross-functional and are profoundly augmented with advanced tools. He sums it up rather succinctly: “Developers are working on more, faster.”
Tsavo goes on to explain how his role as a developer has changed in the past two to three years. Previously, he was able to focus exclusively on writing code in a few languages. Since the emergence of AI systems, he is now able to write code for every team in his company. He says that he now spends less time intensely focused on one particular area or language and now moves faster across the entire organization.
Tsavo outlines some questions developers need to ask themselves amid these changes, including:
- What are users asking for?
- What are investors asking for?
- What are the marketing and product teams looking for?
Tsavo explains how developer teams should determine how the answers to the above questions translate into how they decide what tasks to delegate to an AI system. The result will be divided into a few key groups of functions, according to Tsavo, including:
- Functions that developers can delegate to an AI system
- Tasks that they can do two times faster, augmented by AI
- And functions that they will need to perform themselves.
Tsavo predicts that the result will be that developer workflows become increasingly integrated in what their companies build and why they build it, rather than being isolated and specifically focused on how these systems are built.
Recognizing the Limitations of AI Systems
Tsavo shifts focus to what he sees as the multiple priorities in the development world that are aimed at making overall reductions in three areas:
- Amount of code
- Points of failure
- And security vulnerabilities.
He fully acknowledges that technical debt will continue to be a problem along the way, as it is a problem that AI systems can’t adequately tackle at the moment.
Tsavo then elaborates on the need for world-class engineers working alongside AI systems to double-check and refine infrastructure to make sure all is behaving as intended, explaining that such supervision is the result of how the AI systems are trained. Currently, 60% of code that is represented in AI models straddles ‘the bell curve’ of developer talent, according to Tsavo, and subsequently, this type of code often contains a lot of technical debt and bias. He clarifies the anectodal data on which he’s making these generalizations, “We don’t have, for example, a curated dataset and a trained model on every 10x developer out there. What we do have is a trained model on every developer out there and every kind of codebase out there.”
When asked about what he foresees for software developers given these major shifts in how code is written, Tsavo elaborates that “you’re going to see a lot more of these problem-solvers which are inherently represented in the 10x developer community — where they’re really good at coming in, not understanding anything, quickly coming up to speed and solving really hard problems in creative, abstract ways.”
Though, Tsavo admits that the best models in the world are not at this point yet, and only represents the work of “junior and intermediate developers.” He explains that current developers will end up taking one of two paths:
- There will be developers who quickly upskill to the 10x range and are able to accomplish tasks not yet possible with current AI systems.
- Also, there will be other developers who will not upskill fast enough and have to find new roles, according to Tsavo.
He articulates his belief that, even for highly skilled developers, it is a time-consuming, complex task to refactor poorly written code, often known as “spaghetti code.”
For AI systems, Tsavo acknowledges it would be even more complicated as they would need:
- Extensive contextual understanding
- Enormous number of parameters
- knowledge of how the codebase was written
- The problems it was trying to solve
- Features that were intended to be implemented
The benefit of 10x engineers, according to Tsavo, is that they have all of that information and stakeholders in mind before they build.
Tsavo then explains the dichotomous outcome developers will face: “I think that you either evolve into a 10x developer that complements AI and knows how to use these augmentative tools to move faster in different kinds of environments while problem-solving unique cases all the time. Or you’re going to drop off and maybe work in a different part of the company.”
Tsavo concludes with advice for how developers can use AI tools to complement their work. He describes two questions that developers can ask themselves when faced with solving a technical problem:
- Am I spending most of the time trying to figure out how to solve this?
- If so, can I use generative AI (GenAI) to try to solve this faster?
He goes on to talk about the multiple priorities developers balance in trying to figure out solutions, acquire new skills, read documentation, and communicating effectively with other developers and stakeholders, often simulataneously. “Odds are, generative AI has that information packed into it. You can ask it a couple of questions, and you can be on your way faster. Learning to identify when it is ideal to use these tools is really important,” mentions Tsavo, explaining how developers should evaluate where they are in the process.
If they’re using pencil and paper to develop a specialized algorithm, AI tools will not be able to provide much help in those cases, as certain manual formats are not easily integrated into on-demand GenAI.
However, suppose a developer is working on a problem that does not represent specific domain expertise and is something a multitude of other developers have already worked on. In that case, Tsavo tells Emerj’s executive podcast audience that is an excellent example of when to use a large language model (LLM):
“If you become good at utilizing these tools and identifying when they’re best to be utilized, you’re going to move faster. That’s how you upgrade into that 10x. You know what problems to solve yourself and what problems to outsource to these generative AI models.”
– Tsavo Knott, Co-founder and CEO of Pieces
