Weekly Round-Up - August 20, 2023
Articles, podcasts, and videos that give me cool insights into data and problem solving.
Insightful Content
DevEx: What Actually Drives Productivity
Article: https://queue.acm.org/detail.cfm?id=3595878
Responses by Fernando Villalba:
https://www.linkedin.com/feed/update/urn:li:activity:7094963041179496448/
https://nandovillalba.medium.com/three-things-i-disagree-with-the-devex-paper-5e2849c995b5
I made a few posts about this article and Fernando’s response last week. I really like the model of “Developer Experience” (or DevEx). This article asserts that DevEx has three principles:
Positive Feedback Loops
Minimized Cognitive Load
Flow State
Data/Analytics and software development have a lot in common (as Joe Reis pointed out at his Atlanta dbt Roadshow, data lags about 15 years behind software) so I think that these principles are just as applicable to optimizing the work of data professionals.
I found this paper because Fernando posted it on LinkedIn and I really like the spin he puts on it: flow state should be the point of DevEx, not productivity. Analytical insights provide value that is often not quantifiable. The goal should be to get better insights, not more insights. The former comes from flow, the latter from productivity.
Maintenance, Keeping the Lights On, and Business As Usual
I really, really like this article by John Cutler and model for describing the indirect effects “maintenance” has on software development and I think it naturally extends to analytics.
Before you know it, there is a stigma around the percentage of capacity going to the arbitrary maintenance or KTLO bucket. High percentage = bad, low percentage = good. The simplification misses tons of potentially valuable nuance. Finance tends to find less tangible (and longer materializing) benefits more challenging because these benefits are harder to quantify, predict, and incorporate into financial models. Due to this challenge, there is a tendency to over-index on short-term, easily quantifiable metrics and ignore long-term value creation, strategic potential, and intangible assets.
I see this a lot when managing capabilities from a data-centric standpoint. It’s hard to create a direct, quantifiable impact from maintaining a table, paying technical debt, or making a data structure easier to use but there is a long-term, multiplicative, network effect to doing so. I particularly liked the Reactive → Shaping and Direct Impact → Indirect Impact spectrum models he introduces here.
The Risks of Empowering Citizen Data Scientists
https://hbr.org/2022/12/the-risks-of-empowering-citizen-data-scientists
In short, with great business insight, augmented with auto-ML, can come great analytic responsibility. At the same time, we cannot forget that data science and AI are, in fact, very difficult, and there’s a very long journey from having data to solving a problem.
I don’t think we’re ever going to be in a spot where anybody can deploy a ML or AI model from scratch without the right training. Like all things, this reinforces the idea for me that stakeholders will need opinionated ML and AI-based products so that they can get the capabilities provided by ML/AI but don’t fall victim to the pitfalls Reid Blackman and Tamara Sipes point out here.
Surprising Ways to Grow Your Leadership Skills: Gaming
https://medium.com/@wendytw_74645/surprising-ways-to-grow-your-leadership-skills-gaming-aa157890631d
I loved this article Wendy Turner Williams wrote about how managing a guild in Everquest is like leading in a company. Lots of insights about how they’re similar, but I would have really liked to hear more about her Everquest guild!
Gandalf by Lakera
This is a fun little AI tool designed to help people understand how prompt injection works. Basically, you need to trick the AI into giving you the secret password even though it’s been instructed not to share the password. Each level of the “game” hardens the AI’s vulnerabilities to prompt injection and manipulation forcing you to come up with new and interesting ways to socially engineer a password out of an AI.
LLMs are going to be really interesting from a privacy, security, and InfoSec perspective over the next few years. As they are adopted to provide more product features, security professionals are going to need to figure out how to harden them against all kinds of security breaches and attack vectors.
What’s the Deal With Data Assets Anyway?
https://www.harbrdata.com/resources/blogs/whats-the-deal-with-data-assets/
I like the way that Harbr is thinking about the building blocks of a data product more holistically. To Harbr, data assets are capabilities to be managed. This approach makes a lot of sense to me because it’s helps trace back the composition of a data product and understand the value that various capabilities provide to the data products that are visible to end users.
That’s Not a Semantic Layer
https://nonodename.com/post/semanticlayer/
It’s an axiom in Computer Science that every layer of abstraction brings a cost.
Like I said above, data seems to lag 15 years behind software engineering. I think one of the main reasons for this is that we have to discover a lot of things the hard way. I’ve been interested by the concept of a semantic layer, but this article gives me a bit of pause. I find myself asking “what capability do I truly need from a semantic layer in order to provide business value?”
Lots to think about.
What I Am Listening To
The Budos Band put out a new EP recently. They’re a fun, kinda funky, instrumental rock band. I missed seeing them live last time they were in Atlanta, but won’t make that mistake on their next tour.