Thoughts from JUXT's XT26 Conference - Part 2
Final Summary / Thoughts
Once again I am reminded how great it is to "step back and look at the bigger picture" every once in a while. XT26 gave me the opportunity to do that. It's so great to extract yourself from the day to day and hear the thoughts of such great thought leaders all in the same room.
It was also fantastic to reconnect with old friends and colleagues.
Thanks so much to JUXT for putting this on - I had high expectations and was not disappointed!
Platform as a Product in a 300-Year-Old Bank - Abby Bangser | Syntasso | Founding Principal Engineer - Joel King | NatWest Group | Principal Engineer
A good talk which referenced the Cloud Native Maturity Model which I was not aware of.
- Centralisation: safety through tickets
- De-Centralization - Speed through freedom.
My friend Anshul asked a good question about billing (which I struggle to remember exactly now) but their response was that if you're worrying about that then you must be quite mature in your cloud adoption!
Why Coding Agents Fail to Boost Productivity - Nik Tkachev | JetBrains
Nik explained that:
- his team is somewhere between the two extremes of optimism and pessimism on AI.
- 69% of people agree that AI has increased their productivity.
- But there are interesting impacts at all the following levels....
Individual
- Mental fatigue/Stress - lots of decision making
- Multiple agents running in parallel
- Addictive "You're absolutely right" - makes you feel good.
- I am starting to witness this for myself!
- Fear
- Driven from the worry of "why am I not 10x more productive"?
Team
- Bidirectional pressure
- Expected to deliver more, faster and with minimal additional cost
Other good points
- Just shipping more code isn't good.
- Nor everything in your backlog is worth doing
- The more code you have the more risk - Code is a liability
- I totally agree with this!
Extracting Reliable Software from LLM loops
River Keefer | Antithesis | Principal Engineer
River made the point that LLMs will fill in gaps in the spec with assumptions (this was Henry's point from his talk on Allium - see here).
To get around this problem and to also increase test quality we can make these assumptions explicit by defining property based tests (PBTs).
We can even get LLMs to create PBTs which are then read later for extending the code base.
River introduced a Python framework for creating PBTs which I think was: https://pypi.org/project/pbt/ but not 100% sure
A java equivalent also exists: https://jqwik.net/
Expert Panel: How Far Can We Accelerate with AI?
James Brown, Simone Steel, Mike Jones & Farzad Pezeshkpour, Panel host - Kris Jenkins
Some points from this debate listed below:
- James Brown: Too early to tell what AI is doing for us. We are keeping an eye on our metrics. DORA metrics . James also made the point that we should resist the temptation to find other metrics that show a good story when things are going badly!
- Fuzz: Shared a theory on how organisations adapt is like how empires rise and fall. Two extremes are to be avoided:
- Too much centralised control
- Too little
- Both of these result in the empire/organisation failing
 |
| Photo showing the expert speakers on stage |
Dark Modules, Cobots, and Architecting for AI
Sam Newman | Independent | Author, Building Microservices
Sam made some points in his talk, including:
- We are in a storm - if someone knows "what's going on" they are trying to sell you something.
- We were promised a 10x uplift in productivity, perhaps more like 10%
- This isn't nothing but it's still significant.
- We seem to have less time - people now working longer hours :(
- PRs getting bigger and more frequent .
- Is all code created equal?
- You already trust 3rd party libraries and don't look at every line of code - yet it still runs in your system.
Take time to build trust in AIsad
Sam Newman on stage with one of his slides: Re-examine why you have human reviews.
Comments
Post a Comment