Deep: AI Coding Products Explored
Why AI coding tools are causing developers to quit. How AI coding products can boost development speed. Key features and real world examples. A non-engineering guide to AI coding tools.
🔒DoP Deep goes deeper into the concepts and ideas that are covered in the Weekly Briefing to help you learn lessons from the experiences of top tech companies. New reports added every month.
This week, a FAANG software engineer shared his story about why he has decided to quit his lucrative job in big tech.
It’s a piece that touches upon various reasons why he made his decision, but ultimately it comes down to his belief that the adoption of AI coding products means his job “will be automated by the end of 2025”.
He outlines 3 knock-on effects of adopting AI coding tools:
Managerial expectations can increase faster than AI coding agents can get better
This means that engineering jobs will increasingly be rate limited not by code-writing but by infrastructure management, documentation writing/AI context management, and testing. And, most frustratingly of all, talking to non-technical people
A lot more people are capable of doing at least a semi-decent job of the paste spec → plug into coding agent → test loop than the traditional loop that also included writing all the code
In many ways, he’s right of course.
As AI makes software development easier, managers raise their expectations about what’s deliverable. Roadmaps get bigger, demos get more impressive and release cadences speed up. AI-generated code means product teams will ultimately ship more features but one of the big risks of faster development times is feature bloat; in a strange way, time constraints were a helpful way of limiting the scope of a product and its features.
And while it’s very likely that the adoption of AI coding tools will be a net positive over time, in the short term as product teams try to figure out how best to deploy them, a little turbulence is to be expected. Developers who love the craft of hand writing code may decide they simply don’t want to evolve from doers into delegators - and as we discussed last week, the composition of product teams is likely to change dramatically due to the rise of “vibe coding” and AI code tools.
None of these reservations are stopping AI code adoption rates and startup fundraising.
Over 25% of Google’s code is now AI generated, over 90% of engineers say they use AI coding in their software development process and this week, Anthropic’s CEO predicted that within 12 months, nearly all code could be AI generated.
In this deep dive, we’re going to explore what exactly these AI coding tools do - and what impact they might have on the speed of software development. We’ll dig deep into the AI coding tools from the major players like GitHub and Amazon as well as the various new startups such as Tabnine, Cursor, Codeium and others who have raised tens of billions of dollars in recent funding rounds, to understand what these products do and how their features can make the engineering process faster. And we’ll do all this from the perspective of a non-engineer.
So if you’re keen to get an understanding of what the latest set of AI coding tools actually do — and how they might help you to speed up your development process — this deep dive should help.
Coming up:
What AI coding tools do - core features and differentiators
How leading companies like Instacart are seeing "at least 2x improvement" in development speed by using these AI coding tools
Unpacking the emerging shift from engineers as "doers" to "delegators" as GitHub's new Project Padawan will allow you to assign tickets directly to AI agents
How pricing wars are heating up with Google making a strategic move to offer Gemini Code Assistant for free with a massive 180,000 monthly completions
The 15 leading AI coding products full
What’s included in this deep dive report
Here’s a snapshot of the AI coding tools tools covered in this deep dive:
Product - this includes a mix of established products from big tech companies and new startups like Codeium, Tabnine, Devin, Cursor and others.
Core features - a breakdown of the core features worth knowing about. More on that below.
How it speeds up development - perhaps the most critical part - how exactly this feature speeds up development times for product teams. For each example I’ve tried to make it representative of the unique features that product offers but there is also some overlap across products
Real world example - how product teams are using this in the real world with case studies on their usage from top companies like Uber, Instacart and others.
Pricing plans - Google recently changed its Gemini Code Assistant pricing to make it available for free. A careful strategic decision on their part but most companies charge for access to their AI coding tools. We break down the pricing models for each.
Link to more information to find out more.
Core feature breakdown
There are over 12 different feature categories for each company in this deep dive. They include:
Chat - the ability to chat and ask questions directly with the code base.
Autocomplete - code autocompletes as it is typed based on pre-existing components or conventions.
IDE integration - Integrated Development Environments are where code is written. Most AI code tools ship with integrations into the leading IDEs like Visual Studio Code. Some companies have also built their own proprietary IDEs. We’ll explore some examples together.
Multi-language - support for multiple different programming languages is essential to get the most out of these tools.
Auto-generate from Jira - some have the ability to auto-generate code directly from Jira ticket specs, making it a lot easier to build features quickly. One company says this has reduced completion time by 50%. We’ll explore which ones do.
AI agents - agents that can perform tasks independently, such as debugging, testing, and refactoring code.
Pull Request Support - assists with the code review process by generating summaries of changes, suggesting descriptions, and enforcing consistent coding styles.
Natural language to code - translates high-level natural language instructions into functional code, reducing the need for manual coding of boilerplate or repetitive tasks.
Documentation generation - automatically generates or enhances documentation, such as inline comments and project guides, to improve code readability and maintainability.
A deeper look at some of the AI coding products featured
Now let’s take a closer look at some of the AI coding products features - and understand how they can contribute to boosting the velocity of engineering teams.