GenAI Long term Impact on Software Business Product like SAP, Salesforce, ITSM: Role of Product Prompt Book (PPB)

Leadership Guide

Software comes in so many forms; it can be tailor made just for you. Sofware which is common to multiple users, get into some standardization process and form products. However software products come in wide variety, some are meant for Development of other software’s, some are domain specific software like Finacle, Hotel Management, Hospital Management and some are for common solution to all business like HR, Sales, Business Marketing, inventory, resource planning. Task scheduling etc. This article is for these, generally called as COTS (Coming of the shelf) Software Business Products. Example of these is Salesforce, SAP, HRMS systems, ITSM, PMS etc. etc. This software is little less generalized than software like word, PowerPoint, outlook or messenger. They are more focused on a business process or a business division.

Generative AI Challenge

Anyone who is following examples of Generative AI will know Generative AI can develop software from scratch. Most commonly you might have seen some games like snake getting developed in few seconds. There are few larger games that can also get developed. Some people have tried to develop CRM, messenger etc. and tried to demonstrate its capability. The challenge is Can generative AI develop a complete software product of same quality and support from ground up.

Key Leadership question: Why it Matters

There are two major reasons:

First: It changes the narrative of software business product. If these can be developed as custom requirement then they become more of service projects and then the products. The very foundation on which these products are made is shaken. There is too much repeatable effort to make these products. Given the case it can be generated. It can be generated again for each customer.

Second: There is service component on these products which is customization. This is different part for each customer. This service part can also be generated like core product.

What are we waiting for

Simple question comes, what are we waiting for, why have we not started generating Salesforce and SAPs of the world.

Answer lies in the different level of expertise is develop this software. I mean both development and implementation of these software are multiple years program. Let’s list few

1. Almost Complete knowledge of process domain (Example ERP,CRM)

2. Understanding of customer domain and common elements of it.

3. Software development skills

4. Testing skills

5. Mapping process (From Generic solution to specific requirement).

6. User domain and process knowledge clarity.

7. Hardware expectation or cloud solution etc.

8. Implementation process,

9. Leaders guidance and orchestration

10. ….

Lot of people killed in each of these areas come together and realize this solution. Currently so no LLMs are not mature enough to be good at all these. So, this means there is no way we can write a single prompt, make me ERP like SAP and we get software like SAP. It will not be potentially possible in future also.

How GenAI will be able to do it.

There is multiple level the progress is happening. Let’s list a few

1. LLM specialized in domains are coming up.

2. LLM specialized in process are coming.

3. LLM specialized in development are already in place.

4. LLM specialized in DevOps w.r.t clouds are coming.

5. LLM based solution which specializes in testing are coming.

6. LLM based workflow development solutions are coming.

7. LLM specialized in product development life cycle are coming.

8. Organization are finetuning different models with the text generated in their organization. Weeker link

9. Key process - Agentic process which involves multiple models started coming up standard framework. Example MCP

Once these specialized models leave certain maturity level. Agents can develop this large software along with customization in matter of days.

What is missing

Assume we have all the above things, is there anything which is still missing. There are two major things

First: All these software has developed in decades of development effort. It might have tens of team, and each one might have hundreds of sprints. At the end of each sprint, there is review. This review gave critical inputs to correct course or align and move forward. Each module was later enhanced to cater the real needs. As of software developed using LLMs has no concept of intermediate review, Impact analysis and course correction.

Second: Issue in first one leads the same issue in language of Generative AI. We cannot have a single prompt. There has to be series of prompts to achieve the required features. So, much so, we can call it as book of prompts. Book may have multiple chapters and exercises to be done. It can further be series of Prompt book for a solution which can achieve desired outcome. These series are analogues to product versions. Lets call Product Prompt Book (PPB)

Who will develop PPB (Product Prompt Books)

Open-source software group in respective areas are likely to develop these PPB. Once they start reaching certain maturity level, they will start getting evaluated by organizations and commercial business product developers alike. Software solution like Lang Graph and MCP will play key role in execution of these prompt books.

Who will actually use them

To start with organization which are not able to afford the software or if they are able to afford, they are not able to customize. If they are able to customize, it’s become harder to maintain and constantly paying high consulting fees to product vendors.

Why Product Companies are not doing it first

Product companies are currently focused in developing agents which can do the customization using agents. This is also a key happening. This is reducing cost of ownership of respective software.

What timelines we are looking at

Around 5 years, these things likely to be commonplace.

Key Leadership Takeway

Coming up the LLM model which understand the process of your organization will be key element in the entire mix. Not having this will delay the adoption of PPBs. As they will have to rely on manual intervention at every review cycle. This might become very slow and personalized.