Craig Blumenow

Enterprise-grade virtual-shore IT projects.
10X quality, 2X speed, AI autonomation included, fixed-price, pay-on-success.

[Craig Blumenow @CraigBlumenow]

@Devon_Eriksen_ I've never met you, or your father of course. But your writing, irrespective of the content, is inspirational. So he definitely did a good job.

Sorry for your loss.

[Craig Blumenow @CraigBlumenow]

Random thoughts on Applied AI in business, part 4.
I know it because I’m hip-deep in it!

1. Software user interfaces are disappearing. For example, at Langford Hale we’ve autonomated business processes to such a degree that the user interface is now just your inbox, to manage exceptions.
2. Even so, user training is still a thing. In fact, it’s even bigger, because humans need to understand the entire process even if it’s invisible, in order to unblock exceptions and understand statistics/reports.
3. HITL is currently one of the biggest non-negotiables for company controllers, risk managers, IT managers, etc. But as they learn to trust the agents, and especially the harnesses around the agents, HITL will trend sharply down.
4. Agents managing agents is emerging fast. Again, because trust is evolving.
5. Also evolving fast are governance, rulesets, security protocols, and other ways of ensuring trust.
6. Notwithstanding all that trust, there will always be a need for humans. This is because at the end of the day all this stuff is being made for humans to benefit, so humans will aways need to have some input to define or refine what that is.
7. The way this will all play out is: the best people will take on an ever-increasing share of the human work, and the rest of the people will trend to zero. This means that most people will need to find other jobs, where this is not yet the case.
8. For example, at Langford Hale we have agents orchestrating agents, but also (Pareto) humans conducting that orchestration. But this means that we need way fewer humans than before, because each human can now pull much longer levers.
9. This is a natural continuation of the extant trend for the best-of-the-best people to produce most of the results (Pareto principle), just accelerated.
10. Once this is the case everywhere, Elon’s utopia (everything is free, infinite wealth) will be a reality.

[Craig Blumenow @CraigBlumenow]

The next $1T company will indeed be an AI-native services firm.

But the vertical will have none of those traits, because those are too easy to solve to warrant a $1T valuation.

Instead it will have these:
1) Dispersed data and dispersed conversations, ie specifically not document-driven.
2) Dependent on judgement and fuzzy logic, not rules-bound.
3) Not currently verifiable; only verifiable after the fact by agentic analysis, and then probably only statistically.

I'd love to take a bet on this!

[Craig Blumenow @CraigBlumenow]

Random thoughts on Applied AI in business, part 3.
I know it because I’m hip-deep in it!

1. You cannot manage agents as if they were humans. Not because they cannot act “human”, but because they do things too fast to be effectively managed.
2. Speed is its own quality. That is, enough speed automatically translates into quality. For example, if something doesn’t work, just discard it or undo it, and start again, it’ll only take a few minutes, it’s fine.
3. The ability to undo autonomations (fallback) is almost as important as the autonomation itself.
4. Audit trails are becoming perfect. Most are useless and will never be read, but a very few (the disputes) will be gold.
5. Documentation is becoming perfect. Pretty much all of it will never be read; at most, the user will read an AI-generated summary. But that does not mean documentation is useless, because the record matters, especially for AI’s and especially if there is a dispute.
6. Documentation and deliverables that are relevant for agents are fundamentally different to those that are relevant for humans. Think markdown files vs functional specs.
7. The AI’s are generally very good at calling out where they are uncertain and need a HITL. But they are not perfect. A key challenge for Applied AI is finding a way to know with certainty that you have caught all of these, ie not just blindly trusting the model.
8. Adversarial analysis may be one way to catch these, but (a) it’s very costly in tokens, and (b) a disagreement between models is not necessarily an indication of an underlying issue, ie they may both be right.
9. It may not be possible for humans to correct agents, because the human is even more imperfect than the agent. This is a key insight to keep in mind if you are assessing risk.
10. The appropriate level of risk is not zero.

[Craig Blumenow @CraigBlumenow]

Random thoughts on Applied AI in business, part 2.
I know it because I’m hip-deep in it!

1. AI interfaces-for-normies will be a huge growth area, the driver of successive doublings of adoption. Intuitiveness and ease-of-use will win.
2. Almost all AI interfaces right now are this weird mix of patronising, brown-nosing, supplicant, and aloof. It’s off-putting for almost everyone, but the tool is so useful that everyone overlooks it.
3. Some models are able to inject humour into conversations. This resonates hard, because it’s rare. Whoever does most of that will have a real advantage.
4. Some models learn (somewhat) about you, your preferences and foibles. Expanding this will be huge going forward.
5. Careful about biased models in your business. All models are biased to some degree: wokeness, ruthless efficiency, whatever. This bias reflects the biases of the people in charge of the labs. This could not be any other way, it’s inevitable.
6. Their biases will not match yours, which means the models you will have working for you will not match your company’s bias (ethos).
7. But that’s ok, we’ll collectively work it out. We have a case study from the legacy world of work: your employees have their own biases, but you can still form a team (although it’s not easy).
8. Autonomating a business process is not the end; a better autonomation is coming soon, and that will iterate.
9. People have their own personalities. That means they will interact with an autonomation (automated business process) in their own idiosyncratic way. This is another huge frontier for AI usability.
10. Most workers either think that an autonomated business process will make their lives easier, or will cause them to lose their job. Both thoughts are wrong: more likely the worker will either move up the value chain (manage the autonomations) or move sideways into a different job.

[Craig Blumenow @CraigBlumenow]

Random thoughts on Applied AI in business, part 1.
I know it because I’m hip-deep in it!

1. Autonomating a broken process just speeds up frustration.
2. Wide adoption requires normie-friendly UX. Most people are normies, which means most workers are normies.
3. It’s not the model, it’s the harness. Models are amazing at thinking, but the harness is what allows them to actually do what you need them to.
4. “Normie” is not a pejorative. Meet people where they’re at or it won’t work.
5. Most people (90%+) will be doing different jobs in 5 years. Even if it’s the same job, it’ll be a different job.
6. …but they will all still have jobs. This has been the pattern since the industrial revolution started, and LLM’s will not change that.
7. It’s surprisingly hard to gain productivity in most business functions. The solution is to start, learn, and iterate.
8. Most departments have an immune response to AI autonomation (understandably).
9. A piecemeal approach to autonomation will end up being a nightmare.
10. You need to set up a programme to autonomate the whole business. Your competitors already are.

[Craig Blumenow @CraigBlumenow]

It's an interface/UX/intuitiveness issue. There are a gazillion experiments happening right now, and eventually some will succeed -- and then normies will find it compelling. Same as the early days of any new tech: usenet vs www, early cars were super-difficult to own, early PC's were notoriously difficult to use, early phones needed banks of switchboard operators, etc etc. The UX will catch up, and then catch on.

[Craig Blumenow @CraigBlumenow]

That's just how exponential growth works. You can't see the impact, but it's busy doubling and then doubling again under the hood. The all of a sudden it's everywhere.

An example (one among thousands): using your phone to pay for things. Initially it was scary, it was just some nerds doing it, people thought "this will never catch on". And then, apparently almost overnight, it was everywhere. I bet most people reading this don't even carry cash most days.

[Craig Blumenow @CraigBlumenow]

++ @ScottAdamsSays

[vitrupo @vitrupo]

Eric Schmidt says traditional user interfaces are going to go away.

The WIMP model (windows, icons, menus, pull-downs) was built 50 years ago.

In the age of agents, UI becomes ephemeral. Generated on demand, shaped by intent, not layout.

[Craig Blumenow @CraigBlumenow]

@ScottAdamsSays Scott, I'm so sorry. I just want to let you know -- you improved my life immeasurably, even if I did have to play your livestreams non-live, at 1.5 times speed! Never missed one though. And RRN has been gold.

Didn't see this one coming. The simulation moves in mysterious ways.