2023 Year in Review

A look back and a glance into the future.

First off, thanks to you, my readers.

Back in February, I set a goal to publish a weekly newsletter. I’ll admit to a selfish motivation: I love writing and needed an excuse to devote time to it every week. I had no idea if anyone would read this newsletter, but I knew I had something to say. I gauge success on a simple metric: one person derives value from each week’s post—just one person.

From your comments and casual conversations, I consider 2023’s goal complete. It feels right to roll into 2024 with the same goal. Publish every week and write something of value for at least one person.

Check in occasionally and let me know how I am doing.

Please remember to subscribe and share!

A Year of Whats and Whys

The newsletter ranged across topics, and themes blossomed throughout the articles. I focused substantial print time on software product management because that is my foundation topic. How do you build a product team? What measures and metrics help you recognize success? How do you tell your product’s story? Several of these articles involved me thinking out loud as I built up my fractional chief product officer consulting business.

I spent most of the year thinking about and experimenting with generative AI and large language models. I used ChatGPT, and as the year progressed, I frequently delegated rote tasks to my new GPT-4 AI intern. I struggled to make cool images with DALL-E (more on that below). I used AutoGPT and Godmode to write code and build a simple product on a lazy Saturday afternoon. I wrote about the insane economics of building and running AI models and the bubble forming around AI vendors as they rush into the market and venture capitalists sprinting to fund them at outrageous valuations.

People talk about AI’s existential threat as something AI might do to humanity. Still, it feels like people have a solid grasp on what it takes to wreck humankind, so I wrote about how our actions might hasten climate change as we exhaust even more precious resources and mindlessly charge after the promises of bigger and better models on the way to artificial general intelligence. It’s difficult to see how we can sustain the power and water demands of the massive data centers required to train bigger and better AI models.

Speaking of climate change, I live where the effects of climate change are immediate and obvious: the beach. Humans have a devastating impact on the planet. Counteracting that devastation occupies my mind. This year, parts of Southern California experienced a unique double: an intense winter storm–including first-ever blizzard warnings in several counties–and a hurricane just months apart. These rare events now occur annually.

The only solution is to act.

With the help of the San Diego chapter of the Surfrider Foundation, I started a beach cleanup because you can’t climb a mountain unless you first take a step. And, thanks to the support of the local community, California State Parks and Rec approved monthly cleanups at Tamarack and South Ponto State Beaches in 2024! A product manager mindset might be the answer to breaking down climate change into a solvable problem.

Venture firms poured over $55B into security companies, and something is wrong with their investment theses and the economics of the venture funding model because the frequency and cost of breaches continue to rise to catastrophic levels. The identity security market is flooded with well-funded vendors to chaotic effect and is now rapidly consolidating. 

The year’s most popular post was my musings about the Ping Identity - Forgerock acquisition and subsequent consolidation by Thoma Bravo. Several employees of each company bravely tried to convince me how good the combination is for customers, but that assertion simply isn’t true. These transactions happen for the benefit of a small number of people in the PE firms and the company's leadership positions. Don’t take my word for that; ask the folks laid off as their PE masters implement their “rationalization” plans.

I thought I’d revisit three topics and make predictions to wrap up the year. So here’s the year’s final newsletter looking at AI through the lens of ChatGPT and DALL-E, the identity security market, and the state of product management.

Holy AI, Batman!

Large language models, especially GPT-4, are scary good at creatively completing complex tasks. GPT-4 is much better at reasoning than GPT-3.5 and still outpaces all the other foundational models. (Don’t be fooled by Google’s Gemini theatrics and disingenuous demos. The jury is still out on Ultra.) DALL-E 3 is way better than DALL-E 2.

My GPT journey began tentatively. I goofed around with the free version of ChatGPT, found it somewhat underwhelming, and set it aside. In March, OpenAI released GPT-4, and I misunderstood the power of AI until I started using (and paying for) GPT-4 in September.

In May, I wrote an article about asking ChatGPT to guest author an article on succeeding as a software product manager.

In retrospect, I was unkind to ChatGPT. For one, I used GPT-3.5 and asked it to do something it could not–write an article mimicking my voice and style. For another, I expected ChatGPT to share my disposition against corporate speak and cliche but didn’t give that instruction. ChatGPT failed my expectations, if not the capabilities of the model.

I didn’t learn how to utilize ChatGPT–specifically GPT-4–until the fall when I met Ethan Mollick at an event in New York City. If you haven’t heard of Ethan Mollick, I encourage you to read his Substack and check out his talks. He’s an educator on the cutting edge of using AI in the classroom and presents some of the most unbiased analyses of the various models I’ve seen. His creative techniques for using AI stretch the limits of what a model can do and push the boundaries of how humans interact with it.

The best way to learn how to use AI is to use it. Sure, there are a bajillion prompt guides, but forget those and figure it out yourself. (OK, maybe look around and take a little help.) Challenge ChatGPT with CoT (Chain of Thought) exercises, permit creativity, and encourage it to think out of the box.

My image generation struggles are real

If you’ve been reading regularly, you may have noticed I’ve been experimenting quite a bit with DALL-E and incorporating AI-generated images into my articles. Perhaps I am a victim of marketing. I read the simple prompts used to produce the amazing images displayed on

https://labs.openai.com/

and repeatedly tried to coax DALL-E into creating similarly stunning images. I failed over and over.

As it turns out, I just needed to ask GPT-4, which uses DALL-E 3, to make amazing images.

Here’s my attempt to produce an action shot of a skateboarder in the style of a graphic novel using DALL-E 2. I didn’t ask for the gibberish text nor know which part of the instruction set motivated it to include nonsensical text. This image reproduces the concept but utterly lacks the aesthetics I sought. I assumed I didn’t know how to write a good prompt.

Here is the image GPT-4 and DALL-E 3 produced. While it didn’t render the pose I was looking for–the iconic pushing motion in the image above–it is superior in every other way. You can feel the motion, speed, and sensation at the teetering edge of balance. The colors are good, and damn if that kid doesn’t have a great haircut and some sweet Vans!

My fumbling attempts continued throughout the year. Here’s a cute blue monster intended for another article.

The instructions were to have a one-eyed, blue-furred monster covering its ears in the style of the “hear no evil” trope. After nearly a dozen attempts, DALL-E 2 was utterly unable to interpret that instruction and render a single image of a monster putting both hands over two ears. The closest it got was covering both eyes. 

And here’s what DALL-E 3 produced on the first try.

Art is subjective. But the quality of detail in the floorboards, the texture of the monster’s fur, and the expressiveness of the face are far superior to the DALL-E 2 image. I’m astonished a two-sentence natural language prompt produced this image.

For the article about AI and climate change being on a collision course, I asked DALL-E 2 to create an image representing the vast computing power AI models need to perform the feats we’ve trained ourselves to expect by watching science fiction movies. My attempt produced several variations of this debacle.

This picture is objectively terrible and does nothing to portray the resource consumption required to train bigger and better AI models. 

DALL-E 3, on the other hand, produced this. The difference speaks for itself.

DALL-E 2 came out in November 2022, and DALL-E 3 came out in September 2023. The pace of innovation boggles the mind. 

Identity Security Market Predictions

Back in October, I postulated that the identity security market was rushing toward consolidation. The prediction wasn’t bold. The vendor landscape is crowded and confused; market categories overlap, and products are indistinguishable. Venture funding has slowed as the first movers establish their footholds and defend dearly held market share.

Across the ITDR, IPSM, and CIEM market categories, vendors settled on discovery and continuous monitoring, configuration and usage analytics, enforcing the least privilege principle, and threat detection and response as key product capabilities. Most vendors integrate with the same handful of identity providers, IaaS platforms, and SaaS applications to achieve these capabilities. If you’ve spent any time surveying the market, the messaging blurs into an indistinguishable mess. The differences between products are subtle and nuanced, and customers will struggle to understand what makes Vendor X unique from Vendor Y. 

Tenable acquired Ermetic. Cisco acquired Oort. Okta acquired Spera Security this week, effectively plugging a hole that all the ITDR and posture management companies were exploiting. Vendors relying on their Okta integrations will, by the end of 2024, find themselves hanging on for dear life. 

The conditions for consolidation are endemic. 

Thoma Bravo scooped up Ping Identity, ForgeRock, and Sailpoint, fundamentally changing the landscape of core identity providers and seriously denting the notion that the market wants publicly traded identity management companies.

I predict Microsoft will follow suit and pick up one or two vendors in the ITDR, identity security, or SaaS security posture management space. Possibly one of the other larger identity security vendors, like Wiz or Orca, will start snapping up smaller vendors to fill out their platform capabilities. The combinations will continue until the market settles into a small handful of companies with robust product offerings across a broad capability set.

By 2025, the identity security market will have a similar size and shape to the identity and access management market circa 2010: a few prominent vendors holding 70-80% market share and a handful of small vendors scrambling after the scraps.

The State of Product Management

Punditry abounds, suggesting we no longer need product managers. In November, Snap laid off product managers for rather dubious reasons. AirBnB merged their product management and product marketing functions, which isn’t nearly the big deal the internet made it out to be.

I’ve recently had discussions about what to call product managers. What is the difference between a product manager, a product owner, a technical product manager, a product strategist, and a product marketing manager? Agile practices changed how we define these roles. Sadly, most organizations practice poor agile practices and misuse their product managers.

The ambiguity of roles and responsibilities and the emergence of generative AI as an essential product management tool will fundamentally change how product managers work for the better.

What might that look like?

On a Sunday afternoon in June, I asked AutoGPT to help me build a simple product. The product consisted of an API for CRUD (create, read, update, delete) operations on a user database. Later, I added the ability to query a different data store, retrieve user information, map the user attributes to my data model, and add the record to my database. AutoGPT set up an AWS instance, added all the dependencies, and deployed my product. I planned to add a simple UI but never got around to it. I’d already proved the concept.

The whole project took about 4 hours because locally installing AutoGPT thrashed my poor, aging Intel-chip-based Mac. I switched to Godmode, removing the burden on my overwhelmed machine. I asked Godmode to stop and confirm its plan at each step. I wanted to observe how it formulated the plan and reasoned through the steps. I am glad I took that approach because I learned a lot. But letting Godemode run automatically would have saved perhaps two hours.

AutoGPT now includes a little UI frontend, so switching to Godmode wouldn’t be necessary now that I am running a sufficiently powered M2 Mac.

I don’t write code; in an afternoon, I built and deployed a product from scratch.

My epiphany was that I now had the power to prototype products and features using working code without relying on someone to write the code for me. I remember waiting six months for a group of developers to get me a working API on top of a data store.

AI has profound implications for product managers and how they do their jobs. You can run product experiments independently before you write a product strategy, set a roadmap, or create a backlog of epics and stories. Product managers can devote time to higher-order thinking, freeing up precious time by delegating market research and data analysis, having your AI create interesting ways to present your findings, and working together to write memos, press releases, and product briefs. Product managers have a responsibility to focus on outcomes, not activities. Until now, product managers quickly bogged down chasing activities because they had no delegate for rote tasks. 

AI helps product managers focus on outcomes: building the right product to solve a problem for a lot of customers right now.

Imagine the great product manager on your team no longer burdened by minutiae.

I plan to extend this experiment and quantify how much time and cost a product organization can save when product managers use AI tools to quickly design, prototype, validate, and define product requirements. My hypothesis is at least half the time and ⅓ the cost. ChatGPT is helping me figure it out.

Happy New Year!

I feel a deep sense of optimism as 2023 comes to a close. I encourage you to remind yourself that we live in exciting, transformative times. Keep up with the people and industries generating innovation.

Again, thanks for reading. Have a great holiday, and see you next year!