Impact of Generative AI on Engagement Models and Hiring

The giants of software world like Google, Amazon and Microsoft have invested heavily in using AI / ML to accelerate software development.

Microsoft launched GitHub Copilot which uses the OpenAI Codex to suggest code and entire functions in real-time, right from your editor. It offers APIs for different functions which help make software development at least 10X faster. GitHub Copilot X is already in preview where “X” represents a placeholder on what we can expect it to be capable of doing (e.g., “Copilot <for pull requests>“, “Copilot <for security>“). Click this link to see for what the future has in store GitHub Next. As per official documentation from Microsoft, the code generated by GitHub Copilot belongs to the organization using it and they own it completely. There is no claim of ownership that GitHub/Microsoft has on it. GitHub Copilot for Business does not train on customer code.

Amazon has released CodeWhisperer recently. It is a machine learning powered code generator which can generate real working code. Big advantage is accelerated development and increased developer productivity using automated with high degree of accuracy.

OpenAI’ s GPT-3, released in 2020, was the largest language model in the world. It taught itself to perform tasks it had never been trained on and outperformed models that were trained on those tasks. Since then, companies like Google, Microsoft, and Meta have created their own large language models. ChatGPT which is trained using Reinforcement Learning from Human Feedback (RLHF) and was launched in November 2022 has been a revolutionary application of GPT LLM which has made the world take note of capabilities of GPT models. To define this new class of AI, researchers from the Stanford Institute for Human-Centered Artificial Intelligence coined the term “foundation model”. We now also have GPT-4.

Impact on Commercial Engagements

Large number of engagements in IT sector are staffing based where companies provide experienced employees to customers to augment their teams and help them build products or applications. Staffing engagements are based on pre-defined rate card where each employee is charged as per skills and years of experience. I believe with Generative AI solutions available in software development space, it will have a huge impact on how these engagements are modelled.

Let us assume company ABC has staffed 1000 employees across several projects for client XYZ. Assuming a blended rate of 35 USD/hour/employee, this turns out to be a spend of around 6 million USD / year. Most GenAI solutions claim a productivity gain of around 40-50% (Below is from survey responses from GitHub).

This can mean following.

1. We do this work with 600 employees instead of 1000, mark up the rate for each employee (since they are AI enabled now) and leverage these 400 developers on other projects where we have hiring positions opened up. These provides scale for companies to do more with less, provides cost benefits for clients to get same work done at cheaper cost. Given above example, let us change blended rate to 40 USD and staffed number to 600. Assuming here we will use copilot, we consider the licensing cost of 19 USD/month/seat, this comes to 136,800 USD for year. Considering the revised cost, this comes to 4.2 million USD giving a savings of 1.8 million USD to the customer. And additionally, these 400 employees can be deployed on other projects waiting for hiring to happen to start revenue generation immediately. Win-Win situation for everyone?
2. If it's a managed service fixed price engagements where IT companies own end to end delivery, submit a fixed price contract to customers and deliver what is documented in SOW. In these cases, customers do not want to know how many people are deployed to do the job as long as SLAs are met, CSAT is great and work is being done on time. These contracts in a large competitive bid are priced by each organization differently and cost is a major deciding factor most of the times. These will undergo a change in how we factor in use of GenAI into the contract, understand its impact on the commercials and accordingly price the contracts.

Finance and IT companies will need to work out different models, change rate cards, assess impact on the top line and bottom line before we start adopting these solutions in mainstream production delivery.

Impact on Hiring

The hiring process in most organizations involve undertaking tests and interviews. It involves interviewers giving some problem statements to write a code to understand how much the candidate knows when it comes to coding.

Will this process still remain the same? I expect companies will ask candidates to use solutions like GitHub Copilot or AWS Codewhisperer to write code in real time and assess how they are able to use these solutions. What is being tested here is the ability to give the right prompt to get the best code generated in quick time and not the core programming language knowledge itself.

Prompt Engineering skills (the skill of asking the machine the right questions) have become important to be able to make effective use of generative AI. These skills help to better understand the capabilities and limitations of large language models (LLMs) and developers use the skills to design robust and effective prompting techniques that interface with LLMs and other tools. If more people learn “prompt engineering”, AI will be able to produce very relevant and meaningful content that humans will only need to edit somewhat before they can put it to use.

Impact on Training

YouTube has been used for learning and training for many years now. It is one of the most used platforms for anyone to learn something new, troubleshoot issues and seek any other help. With GPT foundation models and pre-built solutions like Copilot, Codewhisperer, Open AI, will this see a change in how we learn new things and debug issues? I firmly believe yes.

Take an example of a developer who quickly needs to write a terraform script to create a resource on one of the clouds and deploy application. Assume this developer has no clue on how to write a terraform script and what security considerations to take care of. With the advent of GenAI solutions, all this developer needs to know is how to explain the requirements in simple english language. Thats all. The likes of ChatGPT, Copilot and Codewhisperer will generate the entire script which in most cases is 100% correct and developer just needs to execute the script with the right keys.

Generative AI is here to stay and change the way software is developed. Its impact needs to be understood in broader terms beyond just the security considerations and accordingly adopt it for mainstream development and testing work.

Look forward to hearing from others and learn from their experiences.