How Do We Make ChatGPT Act Like A Program Manager In The Flesh?
Large Language Models (LLM) and, in particular, Generative AI are already making a huge economic impact in our society. How do we bring that impact to the next level by achieving automated execution?
Key points in this article
- Generative AI has a significant impact on economies but lacks execution abilities.
- Startups need to build unique, automated applications to leverage the full potential of Generative AI.
- Challenges exist in implementing AI as a Program Manager, mainly due to real-time execution and adaptability issues.
- Integration with other corporate systems could help overcome these challenges, but true automation requires real-time interaction and action capabilities.
- “Digital Serendipity” AI is uniquely positioned to overcome these issues by scheduling impromptu interactions, moving towards a calendar-less future.
- The future of Program Management will lie in a combination of Generative AI and other forms of AI, automating execution while improving productivity and work-life balance.
Proliferation of GPT-powered startups
In an incredibly short amount of time, Generative AI went from a cutting edge innovation to a well-established democratized platform play, allowing anyone to leverage its power through APIs, with very low barriers to entry.
As a consequence, many AI software startups have already been founded to implement vertical and horizontal use cases leveraging core generative capabilities.
According to Forbes, there are more than 1000 startups currently using OpenAI’s GPT as their core technology, and many more are expected to appear in the near future.
Who is going to emerge as the winner?
As AI prompting best practices get increasingly diffused through Social Media, the differentiating factors and competitive advantage built on core Generative AI will likely vanish over time, as use cases will be easier and easier to replicate in competing products.
How do you make sure your intelligent product differentiators will be sustainable over time?
Currently, LLMs are amazing at making sense, but do not have execution capabilities. This is expected, as they are provided as PaaS by large industry giants.
What will make the difference is the ability to uniquely act on the “sense” made by Generative AI, in an automated fashion, rather than having a physical person taking action.
For example, in the current state, ChatGPT is capable of boosting the productivity of a Social Media Manager to unprecedented levels. However, it is not capable of “being” the Social Media Manager, unless someone builds a proprietary defensible application that, among other technologies, also leverages GPT to completely automate the Social Media Management process. This means the application generates plans and contents via GPT, but then it takes action according to the unique “secret sauce” the startup builds.
How replicable is the secret sauce?
My gut feeling is that GPT + a workflow of internal steps + standard API calls to third-party applications won’t make the cut: it is easily replicable and not defensible.
Startups must go beyond that and use various forms of proprietary AI to complement Generative AI.
Program Management challenge
Let’s illustrate the concept by drilling down into a use case example that can be applied to any company and industry.
Imagine you want to build a Virtual Program (or Project) Manager (VPM) application, using Generative AI as the component to make sense of the organization and project context.
ChatGPT 4.0 is really good at taking the following information as input:
- List of team members with respective skills.
- Program/Project description.
- Program/Project goals.
And, based on that input, generating a program/project plan, including:
- Work Breakdown Structure (WBS)
- Identification of tasks with Level of Effort (LoE)
- Task assignment to team members
- Identification of activity dependencies and critical paths
This is pure gold for a Program/Project Manager (PM), who can now save a week of work to build a plan, while having more time to communicate, chase people and schedule meetings to get things done.
However, when you ask ChatGPT to execute the plan, this is what it answers:
As an AI, I can help facilitate the project management process by creating and sharing the project schedule, providing reminders for tasks, and supporting with regular progress updates based on information provided to me. However, as an AI, I can’t actually execute the plan in real-time or ensure that human team members complete their tasks on time.
For example, I can provide reminders for deadlines, but I can’t make sure that tasks are being done as I can’t interact with humans in real-time, or track their work beyond the information that’s provided to me. To keep the project on track, it would be helpful if team members regularly update their progress.
In a real-life scenario, human project managers play a crucial role in driving the project forward, coordinating with the team members, resolving any issues, and making adjustments to the project plan as necessary. They would hold regular meetings to check progress, understand any blockers, and keep the project on track.
So, while I can provide project management advice, guidance, and help with planning, actual execution would still need to be managed by a human project manager.
I am convinced that ChatGPT 5.0 alone will do a much better job at creating the above paragraph, however, it won’t likely be able to execute the plan on its own, without physical or virtual “limbs” attached to it.
You could think about generative models as the brain where cognition happens; your application as the body scaffolding that holds everything together; a series of connectors to enterprise tools and APIs that represent the sensing organs and limbs needed to get closer and closer to autonomous execution.
Since we are reasoning in software terms, I am excluding hardware sensors and robotic limbs on purpose. In the end, fully functioning robots are already a reality and represent a different market with high barriers to entry.
AI + Generative AI = opportunity to sustain a differentiating execution
While Microsoft embeds GPT capabilities into the Office suite and other products, to make their collaboration portfolio even stickier and monetize huge productivity gains based on number of users, only a few companies seem to be working on avoiding having physical employees (or users) in the first place.
You will need a Program Manager on payroll, or breakdown and distribute PM duties over other roles, until your AI solution will be able to fully execute in lieu of a physical person.
We already said that simple workflows aren’t enough to establish and defend a long-lasting differentiation and competitive advantage, so, what do we need to do to achieve such a goal?
Let’s build an octopus!
What if we add intelligence to limbs and sensing organs, to complement the central nervous system?
At first glance, more secret-sauce AI to complement the widely available generative AI might sound like a Frankenstein solution, however, actionable intelligence must be added to cognition, and “more brain” is needed to achieve a defensible level of execution beyond “if this then that”.
The importance of real-life user context
If you look at the previous paragraph where the ChatGPT answer is reported, it looks the limitations that prevent ChatGPT to substitute a PM in the flesh are the following:
- Inability to check the progress
- Inability to understand problems
- Inability to adapt the plan based on updated context
- Inability to interact with team members in real time
- Inability to make sure team members execute their tasks on time and keep the project on track
Feeding updates back into the brain
Limitations 1–3 may be overcome by integrating your Virtual Program Management application with corporate tools such as ERP, CRM, HRMS, Project tools and Collaboration platforms. Updates can be fed into GPT, which, in turn, updates the plan based on new information.
This is something relatively easy to implement by endowing your VPM application with a series of connectors to enterprise systems, plus a series of well-designed GPT queries.
Orchestrating real life with AI
Limitations 4 and 5 are much harder to deal with, and constitute a tremendous opportunity to build a defensible intelligence able to interact with team members in real time, thus opening up to the possibility of expediting projects autonomously.
However, what a PM is able to execute today, might not be appropriate for tomorrow. So, there’s still a question to answer:
In the AI era, what does a virtual PM do differently than its traditional in-person version?
Digital Serendipity revolution
A considerable duty of the PM role is organizing and managing various kinds of calendar events. So, the first instinct would be integrating a calendar management AI with your GPT-powered PM application.
Not so fast…
Traditional Calendar AI limitations
It is no news that calendar management tools are not very useful for a few reasons, including the following:
- People manually block their calendar to be able to work on what matters. So, the AI tools don’t have much short-term wiggle room to effectively optimize.
- Automated scheduling and rescheduling events results in a procrastinated and slower collective decision making process. Since there isn’t a PM that raises the urgency of a certain interaction and negotiates an earlier time with the meeting participants, meetings are pushed out in the future, when the likelihood to find available slots in common is higher.
- The typical 30min calendar slot offers a low resolution of the work day that is not in line with the hybrid work emerging need for faster and more frequent communications.
One of the key reasons for having a PM in the flesh is to look at the program, figure out the relative priority of tasks, and negotiate people’s availability, especially along the project critical path, to make sure a certain chain of events happens and the needle moves in line with goals, times and cost. Negotiating time slots is crucial, as the people that matter are already back-to-back all day and work overtime on solo tasks.
The sad reality is that scheduling and rescheduling is painful, despite advancements in automation. In fact, if you had a free slot on your calendar and a conflict to resolve, that free slot would likely be eaten by one of the conflicting events. Since your calendar is already saturated, you would be pushed to catch up on important things overtime, resulting in poorer work-life balance.
The impact on everyone’s personal life is high and, unfortunately, your favorite calendaring AI, as a complement of GPT, is lightyears far from achieving the same result as a PM in the flesh.
A calendarless future?
The next logical question to ask is:
Is the traditional calendar the right framework to support the emerging hybrid workers’ needs?
If calendar is no longer the right productive framework, there must be a better way to satisfy the VPM requirements related to project execution, in particular:
- The ability to interact with team members in real time
- The ability to make sure team members perform their tasks on time and the project is on track
At the same time, the new framework should overcome the limitations of a traditional calendar-centric Program Management approach, offering:
- Higher work day resolution, not constrained by 30-min slots.
- No more nonsense time slot negotiations requiring an in person PM.
- No need to schedule/reschedule/cancel any event, from meetings to focus time.
- Better productivity and, at the same time, work-life balance.
Free from calendar negotiations
What if the PM role suddenly changed and no calendar slot negotiations with team members were needed to make things happen along the project critical path?
Most of the soft skills would be taken out of the equation, leaving room for automation.
Ok… but how do we do that?
As challenging as it may sound, we should detach our team communications from the calendar. If you think about it, it is already happening via asynchronous messaging: you don’t need to schedule meetings to drive a decision via group chat.
However, crafting messages is very time consuming and leads to slower decisions, when compared with quick live conversations.
How many times have you been staring at a Slack or Microsoft Teams channel waiting for the boss to approve an action, after a discussion that lasted a couple of days?
Have you ever thought how much faster that decision could have happened if everyone were in the same room?
Digital Serendipity removes the need for negotiating earlier calendar slots
At Tweelin, we believe that AI will replace the calendar in 3 to 5 years, putting an end to the immense overhead caused by modeling and managing our days in coarse 30 min slots.
As a first step, we have built a Digital Serendipity AI that interacts in real time with people who need to talk to each other, generating the right impromptu interaction at the right time, through the best available channel in common between two users.
In order to do that, the AI leverages all user devices telemetry, as well as calendar availability (busy/free), and learns user communication preferences.
In its current form, Tweelin receives as input the intention of a user to talk to another user (so called “wish to talk”), however, imagine if the necessity for 2 people to talk were assessed by looking at the updated plan generated by GPT.
When your calendar will be empty and replaced by serendipitous AI, your Virtual PM application will look at the plan, determine that 2 or more people should talk before a certain time to move forward with a task along the critical path, and will engage with those people to bring them on a call (or in a room) at the right time.
Sooner than you think, serendipitous interactions will happen without the need for:
- Negotiating any earlier calendar slot.
- Scheduling nonsense.
All the VPM requirements will be met, and all the traditional PM limitations will be overcome.
Your Virtual Program Manager
Complementing GPT or, in general, Generative AI with Digital Serendipity (a defensible proprietary AI) will be the foundation of a new era of Virtual Program Manager applications that will not be easily replicable and will allow startups to establish differentiation and a sustainable competitive advantage.
Leaders will set business goals, unpack them into a set of projects with related detailed plans, and have the VPM take care of the execution automatically, offering, at the same time, better productivity and work-life balance.
