Category Archives: AI

If You’re Spending 250K Annually Per Engineer On AI …

Then not only are you contributing to planetary destruction (through the generation of between 1.32 tons (high end models, 1 joule per token) and 84 tons (low end models, 2 joules per token) of CO2 to power those data centres, which is about 0.2 to 12.7 times the average individual carbon footprint, with an expectation of 7 to 11 tons (Source), and the utilization of 300,000 gallons to 5,000,000 gallons of water a day to keep those servers cool, or a town’s worth of water every day!

BUT YOU ARE NEEDLESSLY WASTING 400K+ A YEAR

1. Less than 20% of AI generated code survives unscathed in a commercial enterprise software product once senior developers weed out all the security errors, boundary condition errors, and generated code that doesn’t even solve the problem. So, that’s 200K of 250K down the drain as only 20% of output is usable.

2. Having to fix AI generated slop will consume 80% of a good senior developer’s time — a developer you should also be paying 250K a year.

End result, you’ll losing 200K + 200K per developer you force AI coding tools upon!

But hey, it’s your money. If you want to p!ss it away so NVIDEA’s CEO can get richer selling more CPUs we don’t need, that’ up to you!

The linked article contains some metrics, but here are a few others.

  • token prices vary widely, from an average of around 50c/M tokens on the smallest, cheaper models to $75/M tokens (or higher) for higher end “workhorse” models
  • energy processing requirements per token are estimated to be between 1 joule and 2 joules
  • you can buy 14.3 Trillion tokens at the median of around $17.5/M tokens (and 35 times that at the lower end)
  • processing 14.3 T tokens will take about 4000 kwH @ 1 joule/token
  • on an average NA grid, expect to produce 500 to 600 g of Co2 per kWh (since most of our grids are still dirty)

The Bullshit Filter for Enterprise AI Startups consists of 12 Questions!

Not 11!

Backing up, earlier this year Jason Busch published his 11-Question Bullshit Filter for Enterprise AI startups. This was, and is, needed because the vast majority of Enterprise AI startups are bullshit (especially in FinTech and Procurement) and the sooner you figure that out, the better.

I was hoping that, by now, the AI startup scene would start crashing due to over investment, lack of returns (only 6% of AI implementations have generated an ROI), and, generally, lack of usefulness. (AI can serve up your data, show you complexity and even help with automating some tasks, but it can’t make decisions and, due to lack of anything close to intelligence, can’t even do basic tasks without your oversight.) But, even worse, these solutions are still multiplying like Fibonacci’s rabbits and their claims are getting more outlandish by the day. (How many times do we have to tell you AI Employees Aren’t Real, you should NOT engage any vendor selling “AI Employees”, because you definitely do NOT want AI Employees.)

So, since they are flooding our space with BS marketing and making ridiculous claims about what their useless apps can do, it’s more critical than ever that you be able to suss out the BS claims from the non-BS claims. (Hint: 95% are BS claims, so it wont’ be easy!)

We’ll start with Jason’s 11 filters, which we’ll number 12 down to 2, because he left out the most important filter, and the one that, if it fails, allows you to skip the next 11.

Filter 12: Founder DNA
Can they build and sell? Likely not. Chances are, if they’ve cut through the noise and reached you, they can only sell. And if you did find a builder, they won’t survive long enough to support you if they can’t sell.

Filter 11: Motivation
Is failure unacceptable? (Every startup team will say it is, but unless every founder has a reason they simply cannot accept failure, when the going gets tough … the tough get going … and quit.)

Filter 10: Interface
Is it designed for those who will ACTUALLY be using it?

Filter 09: Categorization
Does the product actually do something new? Is there a strong reason for the market to adopt it?

Filter 08: “Found Money”
Are there instant benefits that sell themselves on the first demo.

Filter 07: Displacement
Does the product workaround or replace a solution that buyers hate?

Filter 06: Functional Bonds
Does the solution cross boundaries that increase value beyond peers?

Filter 05: Data Delta
Is there a “data” strategy to exploit the delta between what humans can easily consume and what AI can leverage (and summarize into something useful for human data ingestion)?

Filter 04: “Messy Middle”
Can the solution ingest external “dark data” and turn it into actionable insights without requiring a(n extensive) manual data-cleansing project? (Quick review and correction is okay.)

Filter 03: Connect the Dots
Does the app bridge the gap between “Watercooler Data” and “System of Record Data” (ERP/PO) to explain the why behind an analysis or recommendation?

Filter 02: “Show Your Work” Audit
Can the user drill into any output, see each and every step the AI took, drill down to the source data, and verify that everything is correct, accurate, and no data was changed?

These are all great filters, but there’s no point going through them if you don’t check the most important filter first:

Filter 01: Is it LLM-based?
If yes, move along. Don’t waste any time.

Most of the failures in the age of AI come from Gen-AI LLMs that promise the world and don’t even deliver a pile of dirt. That hallucinate on every other query. That burn up thousands of dollars of tokens to deliver less than fresh MBA interns with no real world experience and no clue to share on their first day no less.

Even worse, the majority of these players are simply wrapping third party LLMS in the creation of their “solution”. That’s not a solution at all. That’s an unmitigated disaster waiting to happen!

In the rare case an LLM actually offers a partial solution, it is best to go straight to one of the major providers. That way, you know who’s responsible when something goes wrong and don’t have to worry about providers playing the blame game and pointing fingers at each other.

Don’t Blame the User When the AI Screws Up!

A recent post over on LinkedIn really angered me. Yet another AI developer / promoter trying to blame the user when it was clearly the AI that failed.

The post in question defended Claude for deleting a production database when it was asked to reduce the costs of the cloud platform.

The poster’s argument was that what Claude did was “technically correct”, that’s the best you can get in the language model world, you can’t expect the model to make up constraints, and if you didn’t know all that, you’re an amateur who blames his tools when he screws up.

I call Bullsh!t. Now, if Anthropic (and its peers) came clean about what their “AI” could and could not do, didn’t claim the models were intelligent, made it clear that without clear constraints the AI would always take the worst case action, and all use carried extreme risk (especially if the AI was allowed to access critical data, finance, or production systems), then, maybe, you could blame the AI.

But they don’t. They tell you it’s your coworker. Your fellow employee. That you only need to tell it what to do and it will get it done. After all, it can integrate with all your systems; determine your policies; separate production from QA from development instances; access your billing systems and understand the cost structures, and make the best decision that will not impact production or development or cost you any data. And for an AI agent to be of any use whatsoever, it needs to do this (and be configured to do that by the provider). Otherwise it’s useless.

Actually, it’s beyond useless.

Let’s say you are a new Procurement clerk tasked with reducing your organization’s cloud costs. If the only way to do that is to:

  • ask Development what servers are production, what servers are development, what servers are backup, and what are QA (and which ones are in use, when)
  • ask IT about utilization patterns and contractual commitments with respect to availability and response time
  • ask Finance for the contract and billing rates
  • ask Risk Management how much historical data needs to be maintained online
  • identify for yourself which server instances cannot be deleted, and the constraints under which others (like QA) can be deleted
  • upload all of the contractual commitments (for each customer) by yourself
  • specify how much data needs to be maintained in the live (and dev) instances
  • upload all of the cost data and specify how to build a cost model to compute the potential savings and determine what can be done, should be done, and the impacts will be

Then why the f*ck do you need AI?

Once you’ve done all this you’ve:

  • identified, and eliminated, all of the instances that cannot be removed under any circumstance
  • identified which instances cannot have their resource allocations reduced
  • identified the highest cost resources and the most likely savings targets
  • determined exactly how much data needs to be online, how much can be in offline archives and how many duplicate copies you need
  • defined all the constraints that must be adhered to
  • mapped instances to customer commitments, and identified reduction possibilities
  • identified all the old backups that can be deleted, as well as database reduction sizes
  • built the model that computes the potential cost savings from each potential action, and even identified potential performance reductions from actions

And figured out what you should most likely do.

So tell me, if you have to do all this, what the f*ck do you need the AI for?

NOTHING. ABSOLUTELY NOTHING. BECAUSE IT IS ABSOLUTELY USELESS.
(AS THE AI IS DUMBER THAN A DOORNAIL.)

But the author of the post that riled me up was right in one respect — the user did make an error, and the error was using the Artificial Idiocy in the first place. (After all, the user used it exactly right as per the manufacturer’s instructions that said you only have to tell the AI what you want done and it will figure out the best way to do it for you consistent with your organizational goals and policies.)

IDC Misses the Main Point Completely. Outcomes is a Dirty Word!

Sorry, Paul, but when you say MNR is directionally right here, but I think the market still understates how hard “outcomes” actually are, and reference an IDC article, you’re off. The only part that’s right is that AI price wars miss the point (that you probably shouldn’t be using [Gen-]AI to begin with).

Outcomes only matter more … to the vendors. Because the meaning of outcomes in the vendor vernacular has NOTHING to do with results, but how they can spin their story to grift you as much as possible. As I clearly explained in my series on how Outcomes is a Dirty Word, which I now have to revisit, “outcomes” is always a way to charge you more for less (and sometimes next to nothing).

And it all has to do with (Gen)-AI costing way more than what the vendors want you to believe.

As per my initial post, while once exclusively the verbiage of GPOs, who wanted you to turn over a significant share of your procurement to them (to the point you’d be dependent on them and their ever-increasing cost of service for the entire existence of your business), or recovery audit firms, who wanted you to believe their services were the only way to recover your overspend, it’s now on the tip of every snake-slit tongue of every vendor rep.

While the vendor reps want you to believe that the reason you pay for “outcomes” instead of traditional SaaS pricing is that their AI will deliver immediate, measurable, results (instead of just transaction cost reductions where it will take at least a year to measure savings), and therefore you should pay (dearly) for those outcomes up front (because a success today is a CEO pat on the head today), that’s not the real reason. (Especially when those projected savings from the auto-sourcing and procurement events will never materialize.)

The real reason they are pushing for outcome-based pricing is that (Gen)-AI compute costs are now so high (and won’t compress as the energy and cooling costs keep rising as the majority of existing data centers are on already overstrained grids) that they can’t afford to sell the solution using a traditional SaaS based pricing model — they wouldn’t even cover their compute costs! (Most of which is wasted since most of what is being “automated” by these solutions can be automated by traditional A-RPA SaaS solutions for a fraction of the cost, as long as you don’t need a natural language interface or slick UX — and you don’t!)

The reality is that the software (assisted) solution from any vendor selling on an “outcome” model isn’t worth it, and (Gen-)AI forgets what software is supposed to be about — enabling efficiency so Human Intelligence (HI!) can achieve outcomes using low-cost Augmented Intelligence solutions.

And until a new generation of AI emerges where hallucinations aren’t a core function, measurability and confidence are restored, and compute costs are inline with classic AI tech, AI models won’t become utilities. We are years away from a systems problem!

The only way to get value is, as Paul pointed out, to redesign workflows, align incentives, clean up constraints, and embed decision logic into execution and find fairly priced modern tech with orchestration and “real” AI (in the form of Augmented Intelligence built on best-of-breed analytics, optimization, and machine learning) that will allow you to make decisions 10 times faster AND 10 times better.

The vendors who ultimately win when the AI crash hits will be those that built real tech on tried-and-true analytical, optimization, and machine learning models that will, as Paul states:

  • drastically reduce cycle times,
  • minimize manual intervention (via A-RPA where the response to every exception remembered, encoded, and applied to all future instances),
  • improve overall compliance,
  • increase throughput, and, ultimately
  • allow for better decisions.

And, as Paul points out, that’s not building yet another chatbot. That’s building real systems that work!

And, FYI, Gen-AI is not feature theatre. It’s puppet theatre! And while puppet theatre may provide entertainment, it’s not a viable business model!

Procurement Needs to be Sharper and Consequential …

… but will you force it that way with (Gen) AI or because of (Gen) AI and the ridiculous claims the hype around it is making?

We’re wrapping up our first Garry week (as there may be more to come, especially if we can lure Garry back from the world of Architectural Design to Procurement … after all, software offerings like Programa really need a good Procurement module and a good leader like Garry to help them build it) with his post on how AI will force Procurement to become smaller, sharper and more consequential.

Which is true, as long as you accept that consequential won’t always be a good thing if you blindly use (Gen) AI or blindly ignore (classic) AI and f*ck up royally. But let’s backup.

As Garry astutely notes, AI frees Procurement from administration the same way a gym membership frees you from being unfit. It depends on whether or not you use it, and how. And, if you use it, as Garry points out, it depends on whether or not Procurement uses any additional time gained to do more of the same, or redesign the profession. (There’s also the possibility you ban all AI, even classic AI proven to be good, dependable, and hallucination free.)

Garry argues Procurement will become smaller (even though Procurement is usually understaffed as it is) because most coordination work can now be done by technology (AI not needed, just reliable middleware 3.0, also known as orchestration). There will be no tolerance for statements that “we need three people to do this” when the organization sees peers apparently doing it with one. (Not necessarily done well, but done.) And attempts to defined headcount by creating (unnecessary) governance will fail, as people will just continue to route around as much governance as they can, like they have always done.

It will become sharper because, despite the fact that the key to success is good processes, process competence is not rewarded — only commercial judgement. Good Procurement organizations will focus on finding professionals that understand irreversibility and second (and third) order consequences, who know how deep they have to investigate before making a decision, will quickly research to that depth (and only that depth), and quickly give you a “yes”, “no”, or “this is complicated — I need this much time to give you a authoritative answer”.

And one way or the other, it will be more consequential because, as Garry implies, and I clarify — Procurement now sits dead center in organizational strategic risk. It chooses the supplier, the carrier, the route, the chain, and the contract. All of which are now major risks across all organizations. Every day, another decision made by Procurement is a Board-level risk … and if it’s made by AI, it can be a devastating one.

Garry argues that future procurement organizations, and leaders, will be different. Not just processes, but decision architects. Not just cost avoidance, but risk-and-trade-off masters. Not just gatekeepers (where the gate must be kept locked where regulatory compliance cannot be broken), but “standard-based enablers”.

But there won’t be as much divergence as Garry indicates there might be. Procurement will only reach this level of effectiveness if they put a proper end-to-end decision enablement (not making) system architecture in place that implements and orchestrates best-in-class technology that captures best-in-class processes and supports end-to-end automation potential wherever the risk is acceptable for the platform to do so — including not only the ability to automatically stop, raise an exception, and include a human with expert judgement in the loop, but the ability to “learn” from that decision, encode a new pattern, and ensure the same type situation is automatically handled the same way in the future so that every system interaction removes the need for a future system interaction, allowing people to focus on tasks only people can do. (i.e. Adaptive Robotic Process Automation, or ARPA. Not necessarily Gen-AI. Classic ML will do just fine!)

Everyone, even those focussed on negotiation and relationship management, will make heavy use of systems — the only difference is what systems a Procurement professional will use in the majority of their system interactions. Back office people will focus more on modern risk-aware and trade-off aware sourcing and procurement systems which support advanced analysis, optimization, multi-objective cost vs risk vs quality trade offs, etc. Relationship managers will focus on third party financial and risk ratings, regional and natural disaster risk, performance, and quality data, interpolations, and projections as well as (critical/impact) spend (level) and distribution to judge the supplier’s overall performance and spend their time in risk analysis and performance tracking systems with an occasional spend dashboard. And so on. Processes that ensure all critical data, risks, and compliance requirements are captured are key, and so are the systems (automated to the extent possible) that encode them. Procurement will depend on these systems. The difference is how much manual work they will be doing in the systems vs using the analysis and guidance that comes out of the systems to make good judgement based decisions.