Suggestions
Ned Lowe
FinTech Builder | Fractional CTOs | Outsourced Teams
Ned Lowe is a technology executive and entrepreneur based in Singapore. He is currently the Chief Technology Officer (CTO) of MISSION+, where he helps visionaries achieve impact by providing principles and people to build great products.2
Prior to his role at MISSION+, Lowe served as the CTO of Singlife, a Singapore-based digital insurer. His career began in the banking sector at Bank of America Merrill Lynch, where he spent 12 years working in equity and fixed income trading technology.2
Driven by entrepreneurial spirit, Lowe ventured into FinTech by founding his own SaaS firm called Argomi. He later joined Amazon Web Services to assist financial institutions with their digital transformation strategies.2
Lowe is also a member of XA Network, an investment network comprising senior leaders from global and regional technology companies.3 He holds an AWS Certified Solutions Architect - Associate certification, which he obtained in June 2018.1
Throughout his career, Lowe has demonstrated a passion for designing and building products, as well as creating teams that share his enthusiasm for technology and innovation.1
Highlights
I attended the Snowflake World Tour Singapore yesterday, and had an interesting epiphany.
One of my favourite articles is Shawn Wang’s “Eating the Cloud from Outside In” (https://lnkd.in/gyq-CW6E), which talks about how Cloudflare ‘surrounded’ the opposition (AWS et al) through incremental, complementary offerings (Cloudflare Workers, R2 object storage, etc) and then won the customer.
It is clear that Snowflake is trying the same strategy - it’s simply not a database provider any more, it’s a cloud platform.
I like to picture data having mass, and in turn gravity - the larger the dataset, the more it pulls other data to it, along with the applications to process it.
However, we often try to push data to applications, rather than allow the applications to run on the (static) data. This is what leads to fragmentation, data lineage issues, and subtle differences in representation.
So how do we get the applications to connect to the data?
Snowflake Container Services allow you to run your containers 𝐢𝐧𝐬𝐢𝐝𝐞 Snowflake. That’s a pure compute play, and a direct competitor to AWS or Azure hosted Kubernetes. Snowpark allows you to run Spark jobs (SnowPARK, Spark, geddit?).
Snowflake Snowsight is basically hosted Jupyter with Streamlit components. Again, running directly on your data - no infrastructure needed.
If my guess is correct - and the economics work out - then they’re probably not that far away from having an “all-in” customer; a company that hosts all their application infrastructure within the Snowflake environment. Needs an API Gateway and probably a better Cloud Function implementation (Snowpark doesn’t count) - but that’s within grasp.
If Cloudflare is aiming to be the fourth major cloud provider, is Snowflake going to be the fifth?
Actually building real applications with GPT/LLMs has been quite instructional, and certainly dampens some of the hype.
I’m a massive believer that the future of technology & teams is going to be fundamentally changed by emergent AI capability, but it isn’t quite so simple. Before we can get complex AI workflows, we need repeatable and consistent output. Getting an LLM to return JSON in a specific format is non-trivial, even with function calling (https://lnkd.in/gtU5UG8N) and other techniques. In many cases, we pass the output to a second LLM to clean it up and force formatting.
I have no doubt this will improve over time, but the idea we can chain multiple calls together into some kind of complex, agentic behaviour is pretty naive right now.
The biggest opportunities I see are not for low value processes that need to be repeated thousands of times a day (where small error rates are magnified), but high value processes which are done infrequently.
At the most extreme: a process that only needs to be done once, but with extremely high value - and ideally we can “dry run” multiple times until we are satisfied. For example, converting a piece of code from a legacy framework to a modern one, with a test harness to prove completeness.
We’re currently building a framework around this stuff to identify which problems make sense to tackle. Let me know if you’re interested in discussing further!
