Stochastic AI Agility: Breaking Cycles of Debt

Author: Miro Wengner

Original post on Foojay: Read More

The launch of ChatGPT in November 2022 has influenced and perhaps even changed the rules of the game in many areas. Although I personally focus on the IT industry, observations suggest that it is affecting the entire industry, including the lives of ordinary people.

The article aims to discuss or address the observed changes at the project management level. Such observations are addressed by rule 17. Stochastic AI Agility. In recent decades, the industry has been intensively trying to implement agile methodologies to iteratively deliver products.

Figure 1. : Agile process schema connected to the AI-LLMs (blue lines and circles) and human knowledge based on experiences (blue bricks)
Figure 1. : Agile process schema connected to the AI-LLMs (blue lines and circles) and human knowledge based on experiences (blue bricks)

The core idea is to have all processes divided into smaller batches that teams try to implement as best as possible or do the best job possible, especially in IT, maybe not only in this regard. During the agile process, teams learn to accurately estimate delivery times, which in my experience is a very difficult task due to many factors. Such factors usually create visible or invisible technical debts that the teams have to deal with, and can lead to various scenarios.

Forbes Technology Council has identified 16 obstacles[1] to a successful software project that affect virtually every chosen development scenario:

  1. Poor collaboration between the product and Engineering teams
  2. Not managing data integrity
  3. Not aligning early on the ‘must-haves’
  4. Overlooking nonfunctional requirements
  5. An Unintuitive UI
  6. Unexpected Complexities
  7. Missed Deadlines
  8. Understanding the time needed
  9. Scope creep
  10. An undefined project scope
  11. Unclear or undefined client expectations
  12. Overlooking speed, security or the UX
  13. Security as an afterthought
  14. Hyper-focused planning and design
  15. Undisciplined backlog grooming
  16. Unclear or incomplete product requirements document

The articles suggest to add one additional point which may be called

  1. Stochastic AI Agility

The name “Stochastic AI Agility” suggests that the output of using AI-LLM definitely contributes to the goal, but the impact may not be exactly predicted. The normal distribution may indicate the level of AI-LLM contribution around the most unknown stages, actually areas of the product.

Figure 2.: Agile Kandban schema with each stages connected to the AI-LLMs (blue lines and circles) and human knowledge based on experiences (blue bricks)
Figure 2.: Agile Kandban schema with each stages connected to the AI-LLMs (blue lines and circles) and human knowledge based on experiences (blue bricks)

The scenarios are later evaluated by management or technical staff with the hope of a better iteration process next time. Yes, we will try to do it better next time and maybe differently.
Figure 3.: Agile Scrum framework schema with each stages connected to the AI-LLMs (blue lines and circles) and human knowledge based on experiences (blue bricks)
Figure 3.: Agile Scrum framework schema with each stages connected to the AI-LLMs (blue lines and circles) and human knowledge based on experiences (blue bricks)

It is a good idea next time. With the development and expansion of the Large Lange Models (LLM) solutions, at basically every stage of development or evolution of anything, things can seem a bit biased, even if it is not visible at first glance.

Figure 4.: Waterfall schema with each stages connected to the AI-LLMs (blue lines and circles) and human knowledge based on experiences (blue bricks)
Figure 4.: Waterfall schema with each stages connected to the AI-LLMs (blue lines and circles) and human knowledge based on experiences (blue bricks)

In my humble opinion, regardless of whether agile (Figure 1,2,3) or non-agile (Figure 4) methodologies are currently used, AI-LLM influences each stage of project development and methodologies. I suggest future research into agile methodologies, as the use of LLM, Vibe-coding, or managerial decision-making based on LLM advice may lead to unsatisfactory results with broad impact.

The point 17, Stochasti AI agility, may be a stochastic process. It may follow a cyclical pattern based on biased recommendations. However, such a fact lowers the probability of success depending on the intensity of the LLM contribution. LLM does not deliver consistent responses, non deterministic.

Each such inconsistency can lead to debts of varying severity. Although it may not be obvious, each project or goal is ultimately a deterministic process with a level of variance, where small batches, bricks, are built up and these lead to the desired goal. It all depends on the level of complexity we are able to focus on to obtain such bricks, small batches (Figure 6.).

Figure 6.: Every debt and challenge can be broken down into small batches, bricks, to achieve a goal.
Figure 6.: Every debt and challenge can be broken down into small batches, bricks, to achieve a goal.

PS: Maybe you feel problems or need to rethink and adjust existing approaches, let’s think about it together and talk about it.

References:
[1] 16 Obstacles To A Successful Software Project (And How To Avoid Them)
[2] Stochastic AI Agility: addressing cycle of debts

The post Stochastic AI Agility: Breaking Cycles of Debt appeared first on foojay.