Future of Code: AI-led Software Development

Hi there!

It’s release day! Tune in here 🎧.

Today, we explore the future of developing software in an AI-led world.

Was this forwarded to you? Subscribe to get the next one in your inbox!

Our world is increasingly written and run in code. Software has created the digital infrastructure powering every industry and has enabled human capital to reach ever increasing productivity. Yet all of the software in the world is created by just a small cohort of developers that represent 0.3% of the global population. What would progress look like if that number grew to 100%?

Jason Warner is one of the most qualified people to talk about software development and how it is transforming with Artificial Intelligence. He believes we have barely seen the start of ‘software eating the world’ and that our ability to produce software is about to explode.

My 5 Year Outlook:

• Everyone will code

• Search replaced by Action

• AI will generate its way out of data scarcity

• Foundation models to become cloud companies

Curious? Read on as I unpack each below 👇🏼

Everyone will code

GitHub’s Co-Pilot, pioneered during Jason’s tenure as CTO, revolutionized how developers interact with code by introducing AI-assisted coding. Today, AI contributes to 40% of code production. While this trend continues, the next significant shift will be moving from developer-led, AI-assisted programming to AI-led, developer-assisted programming. This transition will elevate junior developers and significantly boost software productivity.

However, this advancement benefits only those proficient in coding. More exciting is the prospect of enabling non-developers to code. Enabling the other 99% of the population to build and execute software is a thrilling possibility. With advancements translating code into plain English for documentation, reversing this process to convert English into code is a logical future step.

This shift will democratize software development, dismantling barriers that have historically required specialized training. Imagine a small business owner creating applications to best run their business, a teacher designing new educational tools, an entrepreneur shipping a new marketplace — all without hiring developers. This revolution will accelerate innovation, empower a wide spectrum of individuals, and foster a more inclusive and dynamic technological landscape.

🎧 Listen to our discussion

Search replaced by Action

Beyond democratizing coding, AI’s capabilities can redefine how we interact with technology by replacing traditional search methods with action-oriented requests. Instead of searching for information or tools for us to leverage towards achieving our ultimate goal, we could simply state our end objective, and AI would generate the necessary software on the fly.

For instance, someone needing to analyze sales data could request a specific analysis, and the AI would create a customized program to perform that task instantly. This shift from search to action would streamline workflows, enhance productivity, and provide tailored solutions in real-time.

Jason calls it “full program synthesis”, where a simple prompt in plain english is enough to kick off the software development lifecycle that can be completed end-to-end by the AI model.

It would represent a paradigm shift that will lay down a challenge to search engines like Google as well as traditional software players like Salesforce, which rely on lock-ins with high switching costs. With AI enabling on-demand custom software, the need for expensive, one-size-fits-all solutions diminishes. Users can generate tailored applications, reducing reliance on proprietary systems and long-term contracts. 

AI will generate its way out of data scarcity

Lange Language Models (LLMs) like OpenAI have amazed the world due to the immense amount of data they have been trained on — about 150 trillion gigabytes from the internet. However, improving model performance continuously requires more data, and the quality of the data is crucial.

To address this, OpenAI has partnered with content providers like Reddit, while Elon Musk’s xAI leverages Twitter. These specialized data partnerships are essential and something we will see a lot more of going forward, yet they may not be sufficient alone.

Jason points to synthetic data as a potential solution.

Synthetic data involves generating new data through algorithms that mimic real-world data, overcoming privacy and access issues. This method not only augments existing datasets but also creates diverse, large-scale data for model training. While concerns about diminishing data quality persist, companies like Poolside, which generate substantial original data as their model runs different code executions, demonstrate the potential of synthetic data to unlock new advancements in AI, providing a scalable solution to the data bottleneck.

Foundation models to become cloud companies

As software collapses into the AI model layer and users increasingly come to model companies like Poolside with requests to create software, it opens the question of where will this software be hosted and run? Will the foundation models, who are soliciting the demand for new programs, be best placed to then run those programs for their users and take up the role of cloud providers as well as model companies?

This evolution will likely involve partnerships with existing cloud providers, such as OpenAI and Microsoft Azure or Anthropic and Amazon AWS. This indicates an increasing integration of models and cloud services, pointing to a future where outcomes are requested and delivered end-to-end by a single provider.

This concept isn’t entirely new. We’ve seen it with open-source software like Elastic and MongoDB, where the software itself is open but the hosting, access, and execution are monetized. This business model aligns well with users’ willingness to pay for comprehensive solutions and outcomes.

The future is arriving fast, and with it, an exciting era of integrated AI and cloud services.

Onwards!