The Whiteboard Test: Is Your Business Ready for the AI Revolution?

When frontier AI can solve complex math, how many jobs are truly safe?

Hi there,

When I left my corporate job a few months back, I knew I was going to spend more time experimenting with AI tools.

I don’t think I could have predicted how deep down the rabbit hole I would go.

I’m now much more convinced that the AI-era is not only here, it’s accelerating at a speed that will take many people by surprise over the next few years.

Today I’m going to write about what I’m seeing in AI capability, what that means for business, and what this means for you.

I'll explore this in three parts: current AI capabilities, business implications, and what it means for you personally.

---

Part 1: What can AI really do?

Just yesterday, Google released details of its AlphaEvolve agent. An agent goes beyond just a model to keep iterating through a problem until it’s solved.

Instead of just responding to a prompt, it keeps trying to improve the outcome or arrive at a more efficient solution to a problem.

AlphaEvolve is a coding agent that orchestrates an autonomous pipeline of computations including queries to LLMs, and produces algorithms that address a user specified task. At a high level, the orchestrating procedure is an evolutionary algorithm that gradually develops programs that improve the score on the automated evaluation metrics associated with the task.

The impact of this tool? It helped Google train Gemini more efficiently and solved some frontier math problems and optimised some challenging algorithms.

It’s an example of how at the frontier of AI tooling, novel problems are going to start being solved.

Over at the Meta labs, they released a new model called UMA. The impact of this model could be profound.

New materials and compounds could be discovered faster, and we will learn more about how things work at an atomic level.

I’m not scientist but this sounds cool.

The ability to quickly and accurately compute properties from atomic simulations is critical for advancing a large number of applications in chemistry and materials science including drug discovery, energy storage, and semiconductor manufacturing. To address this need, Meta FAIR presents a family of Universal Models for Atoms (UMA), designed to push the frontier of speed, accuracy, and generalization. UMA models are trained on half a billion unique 3D atomic structures (the largest training runs to date) by compiling data across multiple chemical domains, e.g. molecules, materials, and catalysts.

In simpler terms, UMA helps scientists predict how atoms will behave without having to run costly physical experiments.

This is just the news about AI progress at the frontier in the past week. Of course there is a torrent of AI news and hype. But what can AI really do?

I think the best way to think about this is of AI as a sidekick or problem solving partner.

Any of the leading models (ChatGPT o3, Claude Sonnet 3.7, Gemini Pro 2.5) makes for a decent sidekick if you can guide it in the right direction.

We already know that AI tools can easily do the basics:

• Generate anime pictures

• Summarise an email thread

• Polish a draft of a document

• Find the latest information via web search

• Analyse MRI images or blood work in milliseconds

• Go off and do deep research on a topic and write a comprehensive report

We already know that AI tools are getting better at:

• Code generation across large databases

• Analysing enormous amounts of data at scale

• Acting as a system design and complex technical problem solving partner

• Iterating on problem solving to the point where edge case math problems can be solved or have more efficient solutions obtained

If you’re still in denial about the capabilities of AI and don’t want to spend any money, Google’s AI studio is worth exploring.

You can try out Google’s different models, easily drag and drop documents for analysis and even try out code execution in the browser.

I’ve arrived at the point in my AI journey where it is so blatantly obvious that a lot of knowledge work tasks are better completed with tokens than with human time.

A lot of the “didn’t get the output I wanted” scenarios I’ve experienced are either my own prompting errors or working with data that wasn’t cleaned up first.

This rapidly accelerating level of AI capability doesn’t mean things won’t go wrong.

This doesn’t mean humans still won’t be needed.

This doesn’t mean that one time you asked ChatGPT to write a symphony and were disappointed means that AI is slop.

Yes, we will need safety guardrails and testing and quality assurance and risk management.

As I wrote about last week, we can dial back the catastrophising, but we need to pick up the pace at national leadership level.

It does mean, however, that as a society we need to start thinking a bit more widely about what rapid AI adoption could mean.

Already, I can look at tasks of all levels of complexity performed inside a regulated corporate environment like the ones I’ve spent a career in.

If the latest AI tooling was permitted, an enormous number of people, processes and platforms just wouldn’t be needed.

Part 2: What does that mean for business?

If the latest tools at the frontier, not yet available to the public, are able to contribute in a meaningful way to frontier problems in mathematics and the optimisation of algorithms, how many problems inside a median business have to be solved each day at that level of complexity?

Most businesses operate nowhere near that level of complexity.

Many have never truly embraced technology because they’re small businesses that don’t want to spend the money putting in an ERP platform or the like because they don’t need it.

Some still embrace paper record keeping and go to the bank to deposit cash and cheques. Others have email addresses but not websites.

This is the insight I’m getting from my AI experimentation: most entry level tasks are already best pushed to the machine.

Data entry, scheduling meetings, analysing large amounts of data or generating pro forma reports on a regular schedule can all be arranged accurately with the latest AI tools available at negligible cost.

Few critics share their actual prompts and context in their criticism, until we can easily replicate a lot of this criticism, it’s fair to assume our own skill issues are behind failures to get the outcome from AI that we’re seeking.

Another issue that has emerged from trying all these tools out is the lack of humility: almost every problem with AI tooling I have experienced has been my own skill issue or lack of sufficient context fed to the model or using the wrong tool for the wrong problem.

Each time I’ve slowed down, gone back to the drawing board and worked through clear step-by-step problem solving approaches I’ve landed the result I’m looking for.

But what does this all actually mean if you’re running a business?

Every day you’re busy.

Every day you’re under pressure.

Every day you have a to-do list a mile long.

How can you incorporate AI into your workflows and survive?

Your startup challengers don’t care.

AI-first means before writing one line of code or setting up one process, engaging in a back-and-forth with a frontier model to work through all of the strengths and weaknesses of a business model.

This means literally automating everything, even the boring stuff that often gets left to manual workarounds under the status quo.

The founder will figure out almost all problems with the help of AI and AI agents instead of even bothering to hire a contractor let alone take on the additional overhead of managing a human employee.

But what does it mean for you as a business owner or an executive at a big company? I think increasingly it means you need to find a whiteboard.

On the left hand side of the whiteboard you write down all of the inputs into what you sell.

On the right hand side of the whiteboard you write down all of the outputs you deliver to customers.

What goes in the middle of the whiteboard?

One question - how do I get from left hand side to the right hand side at the lowest level of cost with the highest level of quality?

The answers of how to optimise, enhance, streamline or uplift the middle of a business - its operating model - used to come from consultants or experienced executives doing transformations or putting in new platforms.

The AI era means questioning everything sitting between the inputs and outputs of a firm.

This extends to the physical world in areas like manufacturing - already, factory automation is an enormous category, and AI tools can generate code for CNC machines and the like with ease, it’s not unrealistic that end-to-end manufacturing automation run by AI agents is in our future.

One of the biggest drivers of cloud migration cost blow-outs is when mediocre executives take short cuts: they “lift and shift” processes and platforms into cloud services without re-architecting their operating model to align to the capabilities of the cloud.

The project blows out in cost and delivers monthly bill shock, until someone realises how silly this was and radically simplifies everything so the benefits are actually realised, wasting millions or tens of millions and convincing senior folks that “cloud is bad” when it was just poor decision making and project management.

AI transformation is similar.

You won’t be able to get beyond the entry level savings from using AI more unless deep thinking is done on every possible cost optimisation that can be achieved through embracing AI in each functional area.

Part 3: What does this mean for you?

As the world changes because of this new technology, you’ll face a lot of disruption at work, at school, and at home.

Entire career trajectories will be disrupted and functions that once provided a “career ladder” will end up almost entirely automated.

I’ve written before about building up your own prompts that you use to test different models performance.

The idea of an evaluation is how you assess a model’s output against what you’re expecting.

If you’re thinking about how to deploy AI in your business, building up a library of these evaluations is a good place to start.

But zooming out and up to the much-mocked 30,000ft view is a better first step.

If everything gets automated with AI:

• What is your unique value proposition?

• Where do you play?

• What customers do you serve?

• Are you just a middleman?

This final end state is still a long way away - fax machines are still in use in healthcare and banking - but you need to go forward to the logical conclusion from where this trend is going and reassess what the long-term future of your industry is.

I’m not fully convinced yet on post-labor economics, but I can see where substantial reductions in demand for labour can come from.

Take for example a standard Australian graduate path - becoming an accountant and doing audit work at Deloitte/EY/KPMG/PWC etcetera.

If I train an audit agent on all of the audits I’ve ever done, all of the underlying data, and all of the work papers, and all of the legislation and standards and codes of conduct I have to follow as an auditor, how many people do I really truly need to run that audit capability?

If the cost of running an audit is not about human time but about the cost of tokens being processed by the model, why do I have to restrict my audit to samples of data?

Why can’t my model audit the entire general ledger or the entire process and all of its control evidence?

What sort of people do I need?

I need engineers and domain experts who act like product managers for the agent.

I need salespeople and relationship managers.

I don’t really need to hire hundreds of audit graduates and work them to the bone when literally almost all of that work could be code and done with an agent that connects to client’s systems.

This is clearly a while away, but anything written down and involving keyboard, mouse and monitor use is a target for AI automation.

This means that a lot of existing social status structures are going to come under enormous pressure: what is the point of paying for private school fees and expensive university educations if even fewer people than the already paltry number who make it to the top of each professional field are getting on a “career ladder”?

What if there is no career ladder anymore and essentially all AI is going to do is lock in existing power structures and shut down social mobility through hard work?

The professions have benefited the most from winner-take-all economics, yet they may become the biggest losers, because so much of their tacit knowledge held in their heads from experience, becomes worth much less the moment a reasoning agent figures it out for itself or is trained to do so from the outputs of the current meat-heavy processes.

This means that owning capital becomes more important than human capital - because your experience and qualifications are worthless if you are charging an order of magnitude more than AI unless your value delivered is an order of magnitude more.

This is only going to be applicable for the 1% at the top of each knowledge domain. If there’s no apprenticeship or career track ladder, there’s going to be serious social disruption as entire functions and capabilities are pushed to the machine because people at the top want to take advantage of the AI leverage they can obtain.

As much as Elon Musk’s takeover of Twitter was maligned, the one thing he did in letting go of so many people, was send a serious signal to the global corporate leadership class - the lights can stay on and you can probably cut even more people than you think.

That was a few years ago! Now they all know that AI is here to stay and will face enormous market pressure to deliver higher returns to shareholders regardless of human cost. Do you think the share buyback era happened in a vacuum?

Hyper-capitalism gave us this bounty, and hyper-capitalism is not going to be restrained in the AI era - so buckle up for the ride - you now have a new-additional-permanent-part-time unpaid job where you need to spend time with AI tools daily so you are operating from an accurate fact base when it comes to reasoning about what to do next in this exciting era.

Regards,

Brennan