Clipping
Architecting with LLMs: use plain old code when you can, LLMs when you must
I recently wrote about the Coding agent sandwich , which is an architecture for a coding agent harness. The architecture consists of a tasty filling of LLM invocations sandwiched between two slices of plain old code (POC). Its practical rule is: use POC when you can, LLMs when you must. The following diagram shows how…
microservices.io//post/architecture/2026/09/29/architecti…Architecting for uncertainty - part 2: fast flow defined
This is the second in a series of articles based on my talk Architecting for Fast Flow: Thriving Amid Uncertainty . In this series, I define the term fast flow and explain why it has become an essential software delivery capability. I discuss how fast flow emerges from a socio-technical architecture , which consists…
microservices.io//post/architecture/2026/09/03/architecti…Microservices and GenAI in 2026: my Dear Architects conversation
Earlier this year I was a guest on Luca Mezzalira’s Dear Architects podcast . We talked about where microservices are in 2026 and what GenAI changes. My take is that the best software development practices we had before GenAI - you could even say extreme programming - are even more necessary now, because of the nature…
microservices.io//post/architecture/2026/09/02/dear-archi…Architecting for uncertainty - part 1: thriving in a VUCA world requires frequent deployment
This is the first in a series of articles based on my talk Architecting for Fast Flow: Thriving Amid Uncertainty . In this series, I define the term fast flow and explain why it has become an essential software delivery capability. I discuss how fast flow emerges from a socio-technical architecture , which consists of…
microservices.io//post/architecture/2026/08/25/architecti…Speed limits, GenAI coding agents and Autobahns - part 2: raising the safe speed
Part 1 developed a driving metaphor for GenAI-based development: you are the driver, the coding agent and its harness are the car, and your code and delivery system are the road. And in the same way that the road determines how fast you can drive, the characteristics of the code and the socio-technical architecture…
microservices.io//post/architecture/2026/08/22/speed-limi…Kreps leaves Confluent after nearly 12 years
He announced he is stepping down as CEO, handing over to chief product officer Shaun Clowes. The news follows IBM closing its acquisition of Confluent in March.
sdxcentral.com/news/ibm-in-new-software-blow-as-confluent…Microservices Platforms: When Team Topologies Meets Microservices Patterns - QCONSF 2025 video
The video and transcript of my QCON San Francisco 2025 talk on microservices platforms - the basis for the article series - is now available. Abstract When many teams work on a large, complex application, the microservice architecture potentially enables them to work independently and deliver a continuous stream of…
microservices.io//post/architecture/2026/08/05/microservi…Governance in the age of AI
On the relationship between governance and architecture: procedures that reduce complexity, and targeted checklists that take weight off teams rather than adding bureaucracy.
infoq.com/podcasts/governance-age-ai/Microservices, monoliths and resilient systems
Long conversation on where distribution stops paying off, plus material from the new book. The recurring argument: a lot of people distribute workloads that don't justify it.
youtube.com/watch?v=XV1hf8BwvXcPeople to Watch 2026
Profile on the shift from data at rest to data in continuous motion, and why that turned critical once AI agents started depending on live data rather than snapshots.
hpcwire.com/bigdatawire/people-to-watch-2026-jay-kreps/What nobody tells engineers about becoming a CEO
The full Confluent arc, from a group of engineers with no commercial experience to a public company. The useful part: the difference between what a company can do and what it has to do.
open.spotify.com/episode/3FIAzFqOzfXhdqLGMhes57Eliminating risk with AI
On unified risk decisioning platforms: combining fraud, compliance and credit into one real-time engine instead of point solutions. Familiar ground for anyone who has done high-throughput fraud detection.
oscilar.com/blog/twif-interview-eliminating-risk-aiAI for managing risk in financial services
How institutions process thousands of signals in milliseconds and share intelligence across customer touchpoints. Includes the angle of fighting AI-generated fraud with AI.
oscilar.com/blog/neha-narkhede-barefoot-innovation-podcastThe digital grass isn’t greener. It isn’t grass.
Don't project your constraints onto others who work differently.
architectelevator.com/transformation/digital-grass-greene…Gwen Shapira on technical blogging
Interview on how she approaches technical writing and the community around it, with a round-up of her most-read pieces — including the one on transaction isolation in Postgres.
writethatblog.substack.com/p/gwen-shapira-on-technical-bl…Executive Impact = Logos × Pathos
Engage your audience with facts: from structure to patterns to storyline.
architectelevator.com/architecture/left-right-brain/Invest Your Political Capital
Spend generously but don't go bankrupt.
architectelevator.com/transformation/political-capital/The Mighty Metaphor
Translating your topic into the audience's world allows them to think for themselves.
architectelevator.com/transformation/mighty-metaphor/Autonomous Data Products: The Missing Layer for Data Mesh
Why so many Data Mesh implementations stall at scale even when the principles are accepted. Introduces autonomous data products as the missing layer.
youtube.com/watch?v=PJuNzFK9YaAArchitecture as selling options
January episode where he reframes the architect's job: not deciding early, but buying flexibility. The page aggregates his recent podcast appearances, including the April roundtable with Rebecca Parsons, Barry O'Reilly and Andrew Harmel-Law on designing for chronic dysfunction.
share.snipd.com/person/gregor-hohpe/Wsap3oZCSKeE1noc9lIi0wThe Economics of Technical Speaking
What are 45 minutes of your time actually worth?
architectelevator.com/strategy/economics-technical-speaki…Data 3.0: When AI Outruns Your Data
Her current thesis: AI adapts in milliseconds while the data stack still moves in quarters. Talk on what that forces you to change in the architecture.
youtube.com/watch?v=_j2rMPqv-toWhy systems thinking is essential in tech
The difference between linear and systems thinking, why building a "car boat" is a failure of conceptual integrity, and metacognition as a prerequisite for influencing technical decisions.
techleadjournal.dev/episodes/219/Evolutionary architecture and automated governance
Principles of evolutionary architecture and fitness functions as the mechanism that stops it degrading. Includes her position on choreography over orchestration and on AI's effect on legacy systems.
techleadjournal.dev/episodes/201/Navigating complexity with systems thinking
Conversation with Andrew Harmel-Law on applying systems thinking to software development and to architecture facilitation work.
open.spotify.com/episode/7pQ269A0lvuMTratuBCBnsEnabling Microservice Success
Interview with Sam Newman about her book: out-of-hours support, team autonomy, standardisation, and what she learned from ten years of microservices at the Financial Times.
open.spotify.com/episode/06HZ7FLKIx4qblZNp4UOzEIndex of her writing and talks
She keeps a curated list of her own pieces and talks on GitHub, including the essay on what databases should do and still don't.
github.com/gwenshapChapter 4: Thundering Herds
Open excerpt from the book on the failure pattern that takes ingestion systems down: the herd that arrives all at once after a recovery.
oreilly.com/library/view/building-resilient-distributed/9…Building Resilient Distributed Systems — early access
Book in final editing, with chapters available on the O'Reilly platform. Half technical (timeouts, retries, thundering herds), half on the human and organisational side of resilience.
samnewman.io/books/building-resilient-distributed-systems/
Nothing from this source in the current window. Run the script with a larger --days.