AI Digest
APRIL 2025
'CONCERNS, POLICIES & GUIDELINES FOR AI CODING AGENTS'
Following Reid Hoffman’s posting on LINKEDIN on “AI Writing Code” in MARCH 2025, the foundation would like to issue an AI digest to specifically address this topic. Over the years, companies have competed aggressively to recruit competent programmers (coders) for their organizations, offering high pay, sign-on bonus and stock options. With the advent of AI-driven software development, the use of AI coding agent to write codes automatically has been developed and used nowadays.
With this new capability comes with great benefits but also new concerns. Below, the SAFE AI Foundation highlights and discusses some of these concerns, providing further insights on the need for safety, policies and guidelines:
1. Will AI Coding Agent write most of the code needed by industries?
Human coding is never perfect and hence the need to perform debugging which can be tedious, costly, and time consuming. Hence, it is natural that software industries today would like to automate the process of code writing using AI. This is already happening and the penetration level is below 50% at the moment. Currently, each company makes its own decision as to how many programmers to be replaced as a result of using AI Coding Agents. So far, many are astonished by the capability and the ability of AI Coding Agents. With the fast code writing capability, fewer bugs, and fast learning curve (AI Coding Agents can write code in various programming languages), etc., it is envisaged that AI Coding Agents will be used by software industries from now on and will soon be pervasive.
2. If AI Coding Agents are used, will human coders be no longer needed?
No, human coders will still be needed to manage AI Coding Agents, fine-tune code and ensure overall smooth code production. AI Coding Agents will not only be able to write code, but also integrating parts of code together.
'Human coders will still be needed to manage AI Coding Agents'.
3. If AI Coding Agents are used, will programmers be replaced or retrenched?
AI writing code is inevitable. With the increase use of AI Coding agents, companies will start to lay off human coders or reroute them to other available jobs. A total takeover of human coders will not happen, as no company should run the risk of not having humans to oversee the process and outcome.
'A total takeover of human coders will not happen, as no company should run the risk of not having humans to oversee the process and outcome'.
4. What happens to current lean and agile software engineering processes?
Writing timely, optimized, and efficient code is still an attractive feature for software industries. According to Wikipedia, Agile programming is a “software development methodology that emphasizes iterative development, collaboration, and flexibility, focusing on delivering working software incrementally and adapting to changing requirements” throughout the project lifecycle. The agile process is great for a group of human coders working together in a distributed fashion to correctly build a large code and/or to quickly roll out incremental features needed by clients. However, is this still needed by an AI Coding agent? Do we foresee AI Coding Agents develop code using agile methodology? This seems quite unlikely. The capabilities of an AI Coding agent outperform that of a human coder by many times.
'The capabilities of an AI Coding Agent outperform that of a human coder by many times'.
The way how an AI Coding agent is created is not widely disclosed today due to market sensitivity, competition and the lack of a standard. It is likely that AI Coding agent companies will proceed to sell proprietary software and solutions. Standards for AI Coding agent will take a few years to develop and complete and will require wide industries’ support and endorsement.
5. Are there any Safety Aspects related to AI Coding Agents to consider?
Absolutely. Just as human coders cannot be perfect, AI Coding agents will have their limitations and concerns. Currently, there are no rules, policies, guidelines as to how AI Coding agents should be designed and developed. With no universally accepted method, each company use its own principles and criteria to develop AI Coding agents. The first AI Coding agent will be written by human coders.
'Just as human coders cannot be perfect, AI Coding agents will have their limitations and concerns'.
'The first AI Coding agent will be written by human coders'.
Open questions include:
a. Will an AI Coding agent be given super admin (root) right to modify its own code? If yes, AI Coding agent's code can be modified without human intervention, resulting in risks.
b. Will an AI Coding agent be used to check the code written by another AI agent without human intervention? If so, there are risks involved as it has a “dependency risk”.
c. Can a malicious human coder write malicious AI Coding agents for organizations and later extort organizations for a ransom? This is definitely a possibility. One should not rule out such possibilities. While an AI Coding agent does not understand morals and ethics, it follows exactly what its construct (its creator) tells it to do.
'While an AI Coding agent does not understand morals and ethics, it follows exactly what its construct (its creator) tells it to do'.
d. Which organization is leading the principles, policies and standardization of AI coding agents? As expected by the SAFE AI FOUNDATION, ChatGPT says no one!
e. Is there is a comparative index or benchmarking for AI Coding Agents? Currently, there is no standardized index or benchmarking for AI Coding agents.
6. Final Remarks
AI writing code in inevitable. Industries and our society have to face this reality. This digest has addressed some of the concerns of AI Coding agents and its impact on industries and the associated risks. There is a need for a universally accepted AI coding agent development guideline, best practices, policies to govern safety, accountability and agents management. The foundation advocates AI industries to work together to make AI Coding agents "safe for all" to use.
~ ~ ~ end ~ ~ ~
REFERENCES ON AI CODING AGENTS
1. Reid Hoffman – Will All Code Be done by AI in 2026? https://www.linkedin.com/posts/reidhoffman_ai-coding-applications-have-taken-the-world-activity-7308880112702083072-YBZO?utm_source=share&utm_medium=member_desktop&rcm=ACoAAAEAYQQBINZmDdQn07AYZFpwwk5gicYpewI
2. CodeGPT – AI Agents for Software Development. See: https://codegpt.co/#:~:text=CodeGPT%20is%20your%20go%2Dto,regardless%20of%20size%20or%20host.
3. GitHub - Co-pilot - AI Coding Assistant. See: https://github.com/features/copilot
4. Zencoder - AI Coding Agent. See: https://zencoder.ai/
5. Canva Code - See: https://www.canva.com/ai-code-generator
6. Gemini Code Assist - See: https://codeassist.google/products/business
7. Meta Code Llama - See: https://about.fb.com/news/2023/08/code-llama-ai-for-coding/
Disclaimer: The information in this digest is provided “as it is”, by the SAFE AI FOUNDATION, USA. The use of the information provided here is subject to the user’s own risk, accountability, and responsibility. The SAFE AI FOUNDATION is not responsible for the use of the information by the user.
Notes: The SAFE AI Foundation is a non-profit organization registered in the State of California and it welcomes inputs and feedback from readers and the public. If you have things to add concerning policies, guidelines, and best practices for AI Coding agents or would like to volunteer, please email us at contact@safeaifoundation.com
authored by SAFE AI Foundation
~ ~ \\ // ~ ~
