I gave my 2026 YOW! Tech Leaders Summit talk the alternative title lossy compression, as it squeezes 28 years in tech into 25 minutes. Many individual slides are are pointers to multiple blog posts or book chapters in their own right! This post includes the main references, by section from the source slides.
Applications over three decades of tech
Describe:
- CAD projects from turn of the millennium in Throwback Thursday
- Descriptive spatial data (from a later period) in I did it my way – hand-rolled navigation with open spatial data
Predict:
- Simulation of car crashes and contact centres in Leave Product Development to the Dummies and more posts tagged simulation
- Machine Learning evaluation in Cost-Sensitive Learning – A Hitchiker’s Guide and and example projects in the Lockdown Wheelie Project Part 2, the Slackometer and Nerfing Along
- Graph techniques in Lego and Software – Part Roles as well as I did it my way – hand-rolled navigation with open spatial data
Explore and solve:
- For design exploration see No Smooth Path to Good Design and see also explore examples in posts tagged simulation
- Analytic maths solutions for AR meteorite tracking in A/VR with Horizontal Coordinates in iOS and Android, Seeing Stars – Bespoke AR for Mobiles (YOW! West & Connected 2014) and Fireballs in the Sky
- Numerical exploration and optimisation methods examples in Maths Whimsy, the associated PyCon AU 2023 talk, in particular solving puzzles, and lately the whole trippler EV charge planning series
- Embeddings and generative AI in the Lockdown Wheelie Project Part 3 and A gentle introduction to embeddings, which includes further discussion of human-machine hybrid solution exploration
Many of these are examples of how I maintained technical depth in leadership roles.
Know your demand
On strategy:
On research projects:
- Asymmetries of product development payoff in No Smooth Path to Good Design and Dumbell Delivery; Antifragile Software
- What to expect when you weren’t expecting an R&D project
- R&D burn up
- Staging R&D by risk in An evergreen question: What is an MVP? and further discussion in trippler talks
- Playing Games is Serious Business for R&D stakeholder engagement
- Dual track delivery in Electrifying the world with AI-augmented decision making
On change:
- See (on review, an eclectic collection!) posts tagged change
Specialise together
Cross-functional teams:
- For AI/ML product development: Chapter 2 of Effective Machine Learning Teams
- For internal team dynamics and collaborating across specialisations: Chapter 10 of Effective Machine Learning Teams
Continuous social learning:
- Nonaka cycle introduced in Reasoning About Machine Intuition
- Illustration of human-machine hybrid solution exploration and learning in A gentle introduction to embeddings
- Thoughtworks Data and AI mini-blogs
- Thoughtworks “Shokunin” challenges in Maths Whimsy
Having fun:
- About safetydave.net and a reflection on 10 years of the blog
- Playing Games is Serious Business
- Chapter 10 of Effective Machine Learning Teams discusses fun and playfulness
- Visual puns for nerds
I think there’s potential to write more up on recruiting distinct specialities and the “data thinking” model of balancing end-to-end execution with raw solution power (this demonstrated in product design of trippler and iFlute).
Flow at any scale
Cross-functional teams and rapid iteration:
- Case study of REA Group Automated Valuation Model improvements in partnership with Thoughtworks
- Inter-team costs in Why are teams twelve times faster?
- The new new NPD in Reasoning About Machine Intuition
- Continuous Delivery for Machine Learning covered in chapter 9 of Effective Machine Learning Teams
Scaling patterns:
- Chapter 11 of Effective Machine Learning Teams for inter-team effectiveness
- Spun off into Effective Topologies for Data and ML teams talk
- Collaboration with Matthew Skelton from Team Topologies in Stream of Teams Podcast and Team Topologies x EMLT
Working with the grain of today’s tech
Testing with non-determinism:
- Getting serious about testing AI
- ML evaluation/validation/calibration discussed in LLMs as text simulators
- Simulation evaluation/validation/calibration also discussed in Playing Games is Serious Business and Leave Product Development to the Dummies
- Modularity in solutions tackled in Continuous Intelligence (AI is narrow but composable) and 22 rules of generative AI – architecture and software engineering and lack of tackled in GenAI Stone Soup as well as Antifragile AI Architectures
- Testing covered in chapters 5 & 6 of Effective Machine Learning Teams
- Continuous Delivery for Machine Learning covered in chapter 9 of Effective Machine Learning Teams
LLMs as text comparators:
- See the whole Semantle solver series
- And A gentle introduction to embeddings
- What not to do: LLMs are lineage black holes
From here, some of the content was newly written to help refine the message I wanted to convey in the talk.
LLMs as text simulators:
- LLMs as text simulators
- LLM WTF to dive into the mechanics of next token generation
LLMs as answer search:
- For products that provide answers: Antifragile AI architectures
- Searching for programs: Hard problems in highly agentic coding
Durable core, ephemeral shell:
- As a key pattern in Antifragile AI Architectures
- Reflected in the LEGO reuse lessons from Humour Me – WRY vs DRY
- And also supported by the seclusion and surrender strategies for maintaining responsive development from Scaling Change
Well, I hope you found that useful and can navigate to something specific, rather than being overwhelmed by the volume of content! This has at least been a really great exercise for me to update my graph of connected ideas and content.
Leave a Reply