• MISSION+
  • Posts
  • February 2025 — Reflections On The Mission

February 2025 — Reflections On The Mission

Author

MISSION+
February 05, 2025

It was always going to be a politically charged month with the US inauguration, but I don’t think anyone was expecting the $TRUMP memecoin. I personally bought in relatively early, but for a small amount - I was pretty sure his X account had been hacked! Nope, not a hack - just a confirmation that we live in some kind of soap-opera simulation for the entertainment of our creators.

But even the $TRUMP coin was overshadowed by the DeepSeek R1 release. This freely available model (including weights for self-hosting!) is state of the art, comparable to the best models out of OpenAI.

Much debate now lies on whether they are really a small independent lab that somehow managed to train a benchmark beating model for $5 million dollars, or a front for the Chinese Communist Party - hellbent on destroying the economic value of American AI giants by launching competing models, then making them free. Food for thought.

But it looks like the top new open weight/self-hosted model (according to benchmarks at least) belongs to Alibaba, with Qwen 2.5. All these new models demanded a response, and OpenAI delivered - releasing o3 mini and making it available to all for free.

Somehow along the way, Google’s Gemini 2.0 Flash Thinking was lost amongst the noise. A first-class model, almost comparable to DeepSeek, available for free at Google’s AI Studio.

The game has changed: any organization thinking about self-hosting should at least be considering DeepSeek or Qwen, and anyone cost-conscious for their API calls should be testing the integration with any of these new models.

MISSION+ Update

We recently did a refresh to the MISSION+ web site, highlighting more about our specialist tech leaders, and the principles behind how we like to build products. Take a look and share your feedback, we’d love to hear your thoughts!

Reflections On The Mission

Pascal’s Wager states that one might as well believe in God, as it costs nothing and may save your soul.

The problem (acknowledged but dismissed by Pascal) with this is it assumes only God can rain down eternal damnation. But what about Odin? Or another god that humanity is yet to discover.

To me, this illustrates the issue with judging problems in only one dimension.

The “obvious good”, or the “it costs nothing so why not” options are often not obviously good (nor cost nothing!). But these can be the hardest thing to argue against. “What do you mean you don’t want to reduce risk?!”. “What do you mean you don’t want to future proof?!”. “What do you mean you don’t believe in God?!”.

One of the biggest areas where I see this run rampant is risk & compliance. Too often, risk management becomes risk elimination: even with the best of intentions, it is too hard to argue against the “obvious good”.

This goes beyond just internal policy, but leaks into regulation. I have no doubt that rules have been put in place with the best interests of all in mind. But at least some regulatory decisions seem to have been judged in one dimension without considering trade-offs.

An extreme example of this is around anti-money laundering (AML) regulation. On a single axis, of course we want to reduce crime. But add in the cost of compliance, and the picture becomes more nuanced. A 2020 paper highlighted this - estimating there was less than 0.1 percent impact on criminal finances, but compliance costs exceed recovered criminal funds more than a hundred times over. This means banks, taxpayers and ordinary citizens are penalized more than criminal enterprises.

AML-originated frustration abounds - no more so than with many banks asking their customers why they want to withdraw relatively small amounts of cash. This Australian guy explaining to the bank teller that the money is for the “devil’s lettuce” always makes me chuckle.

A second area where I see this one-dimensional argument is architecting for scale. Who wouldn’t want to be able to support millions of future users?! The problem is that all those microservices you’re building, and all those abstractions for future swapability, are killing your productivity. You’ll never need to utilise that scale, you won’t get there before the product fails.

As measured against the one dimension of scalability, it always seems as though architecting for future scale is a Good Thing. Architects try to balance whether to introduce the complexity now rather than later. “We don’t need microservices yet”. Embedded in this statement is the assumption that we’ll need them some day, and that it’s just a question of when to invest.

Viewed through a more complex set of evaluations, I assert that in many cases it isn’t a question of when, but a question of if. We’re not just deferring the implementation, we’re objecting to it.

The result of this is that rules and guidelines grow over time, with no real balancing function to push back. Every addition “makes sense”, with any objection shot down with the common refrain “what, you don’t want to reduce risk?”.

How do we resist this? The only real option I see is creating an explicit function that is empowered to push back, with their KPI being the amount of rules or procedures removed. Leaving it in the hands of those setting the rules (or being ruled) is too prone to abuse.

I haven’t seen a function like this in the real world, and so I am intrigued to see how the US’s Department Of Government Efficiency (DOGE) works out - it could become a useful precedent, or a cautionary tale for the future. Let’s revisit in a year’s time! If any reader is aware of successful ‘reduction’ functions in a major enterprise, please do share.

Interesting Articles

This Is The Way

Feature A Fractional

We’re On A Mission

Until next month,

Ned & the MISSION+ team

Enjoyed this? Share this with a friend who might too!