“Agile GPT” — Removing the Blank Canvas Problem to Everyday Work

A core component of Lean Startup methodology is the build-measure-learn feedback loop. The first step is figuring out the problem that needs to be solved and then developing a minimum viable product (MVP) to begin the process of learning as quickly as possible. Once the MVP is established, a startup can work on tuning the engine. This will involve measurement and learning and must include actionable metrics that can demonstrate cause and effect question.

The?Lean Startup, By Eric Ries, fundamentally shifted the paradigm for entrepreneurship and the formation of new companies. Specifically, the build-measure-learn feedback loop that is foundational to Lean methodologies requires iterative experimentation to gain more knowledge on the problem set, possible solution set, and acceptability by the end customer. This iterative mindset introduced by The Lean Startup helped set the foundation the?Agile Methodology, adopted widely by Product Management and Engineering team members alike, spanning start-ups to large enterprises.

Previously, I worked at a Digital Transformation and Consultancy company called?TribalScale, whereby I acted as a Transformation Product Manager. My responsibilities included embedding myself directly on product teams within large enterprises, our clients, where I employed a learning-by-doing methodology — whereby I “paired” with the Product Manager on the team and taught?Extreme Programming?and Agile Methodologies to them while directly working on the product. What I gained by working at Tribalscale was not only the consulting and project experience, but also a deep understanding of the essence of being a product manager and the best practices a product manager should employ within the Agile and Extreme Programming Framework. This foundation was further solidified as I had to teach other Product Managers (and also wrote a couple of blogs). One of the largest concepts I taught was concerning the Minimally Viable Product (“MVP”).

The?Minimally Viable Product, by definition, is the minimum set of feature requirements that satisfies a unique problem for a set of customers. The Agile methodology thereafter dictates iterative improvements of the product over time, gathering direct feedback from customers (qualitatively) and via metrics (quantitatively) to build a product roadmap that continuously adds value to the product, as opposed to offering all features at inception and foregoing the learning process and market opportunity along the way.

One particular example sited as many case studies within Business Schools is with GrubHub. GrubHub started from the inkling of an idea, a hypothesis per se, that customers would pay for the delivery of food directly to their doorstep. As the first experiment, the founder went door to door soliciting orders for pizza and would physically drive to the pizza parlor, order the pizza, and deliver it to the customer’s doorstep for a fee. This example of an experiment detailed the problem that customers wished to be fulfilled — simplified food delivery from restaurant to doorstep — and what would eventually be built into GrubHub. This anecdote clearly illustrates the concept of a Minimally Viable Product (MVP) and is meant as a leading example to the core concept to be articulated in this blog post.

My claim: Generative AI capabilities, powered by Large Language Models, can accelerate the creation of?any?Minimally Viable Product or Proof of Concept and remove the blank canvas problem.

Why make such a claim? Let me explain.

Van Gogh stated:

You don’t know how paralyzing that is, that stare of a blank canvas, which says to the painter, ‘You can’t do a thing’. […] but the blank canvas is afraid of the real, passionate painter who dares and who has broken the spell of `you can’t’ once and for all.

The above illustrates a concept known as the blank canvas problem — that sheer paralyzing fear of starting from scratch. Arguably, the start to any project — writing a product requirement document, drafting a preapproval letter template, programming an application, crafting marketing copy for a new web page, building a cover letter — can be daunting. Detailed and meticulous planning can take place prior, but even the sharpest of minds experience hesitation when starting from scratch.

Enter?OpenAI?and the?General Purpose Transformer (GPT). The internet is clamoring about how GPT-based capabilities and Generative AI will nullify many jobs in the industry today. Existing also are viral examples of hallucinations, proliferation of misinformation, and lack of transparency on what content has been produced by AI versus a human. This blog, while acknowledges the shortcomings of Generative AI and general skepticism of the technology, is meant to highlight where Generative AI can?assist?humans, not replace them, and?draft?content, not publish.

Take many of the aforementioned examples: (1) Crafting a cover letter, (2) Writing a Product Requirement Document, (3) Building a preapproval template for a mortgage, (4) crafting marketing material for a webpage. By virtue of Large Languages models being trained on the broader internet at large (generally speaking), these models have probably observed many instances of the aforementioned documents, understand the basic structure of them, and have vectorized common terminology and associated verbiage associated with them. Below are some of the examples generated:

Returning to the argument that GPT based functionality will nullify our jobs, I argue that GPT related capabilities will accelerate human capabilities and make us more efficient in our day-to-day jobs. A common quote I keep hearing circulating the internet is “AI will not replace our jobs, rather, someone working alongside AI will replace our jobs” and I think that claim is exactly right. Taking the above tasks as an example, what “temperature” configurations and associated generative text thereafter can not account for in the drafts being created by Generative AI is precise diction. Whether describing yourself in a cover letter, the syntactical structure of a legal sentence, or value-based language in a marketing website, diction should be highly precise and every word matter.

Take the above examples. ChatGPT generated generic templates for a mortgage preapproval and cover letter. Exploratory engineering projects that I’ve heard of seek to parameterize these templates to make prompt engineering more user friendly. Nonetheless, the temples to do not take into consideration bank specifics or language incorporated from the job description, but whose to say that could not be inserted into the prompt or build using few-shot learning?

Language is inherently nuanced as well as highly powerful. The specific words chosen in a particular message can invoke specific intended emotions, solicit specific connotation of sentiment, and definitionally be more in tune to the message trying to be conveyed. Personally speaking, I’ve tailored my bio to reflect precisely the type of personal imagery I seek to portray. Using me as an example, I could utilize GPT to generate a base bio with some of my notable achievements within the prompt definition along with other important parameters, generate a sample bio, and then carefully tune the diction to my satisfaction, thus resolving the blank canvas problem and accelerating the time I could write a self-bio and iterate language over time.

Going back to the Lean Start Up, one of the primary objectives of any experiment is to optimize learning for the cost, cost being financial or temporal.?The user of Generative AI in such a scenario to quickly generate an MVP of any task, from cover letters to applications, can rapidly decrease the time to learning and accelerate the development process of any task.?The above use case is fully cognizant of the creator disruption movement happening in response to Generative AI but can still largely apply (albeit the legal response that needs to solidify). As a Product Manager, I will seek to apply this methodology to everyday tasks within my career both personally and professionally, with transparency and I encourage you to do so as well!