Art credit @Madison Harney
Token economics have entered the chat
In mid 2025 I began to worry about how we were accounting for AI spend. I asked people who are smarter than me on this topic and got deeply unsatisfying responses. This is when I knew things were going to get awkward.
Having run large scale operations with massive margin improvement goals, I understood instinctively that the math would become uncomfortable quickly. When you run massive organizations you realize that small "perks" become expensive. Giving everyone a t-shirt or a spot bonus or even a clutch patagonia jacket at a startup is an easy decision. Doing that for an org 10x the size might make you think and for a group 100x you start to feel pretty uncomfortable, especially when you recognize that budget miss could result in layoffs to reconcile the gap. You have to ask yourself is giving everyone a t-shirt the right call in a tough budget year? Maybe it is, but maybe not. Maybe you decide it isn't and the layoffs happen anyway.
If you follow the tokenmaxxing stories you notice that people are beginning to see what I was concerned about. Side note, this is a recurring pattern for me, I've learned that most people don't understand what I'm talking about until the problem is real for them. It makes me empathize with Cassandra.
With each phase of #AI transformation, I have been uncomfortable in ways that I struggle to articulate. I was able to write a note about the early days of AI Fluency programs and I have seen some good discussion on the obvious challenges of token leaderboards. The challenge for me goes a bit past Goodhart's law, but that is a solid place to start. Measuring a thing is good and also a risk, especially when the "why" gets lost along the way (as is VERY apt to happen in large groups, and with people who have been conditioned to expect a playbook in lieu of first principles understanding).
I lifted this from Twitter/X- can't remember where for attribution - apologies.
So instead of just living in the world of I told you so, I want to offer a few suggestions on how to break down this problem. I am using a financial tracking lens because I believe this is the missing piece right now, not because this is the only way to look at opportunity. I also want to remind everyone that the opportunity is new value capture. Capital allocation should be in service of that goal.
Given my intention is to Invent the future, I need more people to be thinking about sustainable innovation vs. the current behaviors (YOLO'ing tokens and hoping something smart emerges).
Financial tracking
You need to be tracking both usage and budget in the following buckets. It's going to be hard to get the specificity you need - instead of waiting for precision expect to refine as you go, especially for those costs that have to be allocated vs. attributed. Take a guess and document your assumptions - refine over time but whatever you do get started now.
Cost of Sales - break down AI that you use to generate revenue. This is both what is used in the delivery of your product AND what is used for customer acquisition. Keep track of tools, FDE and tokens and evaluate unit economics and margin closely. You are likely to observe that your intuition for how your business works and what "good business" looks like will change dramatically in the next few years lean into that learning. Ask a lot of new questions to get grounded in the numbers.
Internal IT Costs - The cost of worker productivity is also changing. There is a subset of the AI costs that looks a lot like email or laptops or cellphones. Things that are non-negotiable for work and generally beneficial to the business in broad ways. You don't monitor who can use excel today- you trust your workforce to use excel if they need it. Microsoft is counting on this (and Google of course) in their growth plans and this general spend should not be confused with the bigger value prop of AI. This should be where you capture your existing vision of "30% productivity improvement".
AI Initiatives - here is the big one and where there a lot of confusion is living. I think many businesses are beginning to realize that not all AI projects are worth the cost and yet sorting that out when both cost structures and organizational priorities are evolving quickly is very hard. Here is the place where most businesses need to put on their transformation hats and ask the bigger questions - things that used to live in the build/buy/partner, business transformation and [portfolio/business/product] Strategy domains. A few things to think about in this bucket:
Pick a small number of specific Initiatives - stop with the thousand flowers nonsense, you don't have time for that anymore. It is the job of leadership to pick the right initiatives that will be the most impactful for your business goals. It is ok to lack certainty and it's totally fine to adjust this over time, but getting moving on something specific and strategic will be more helpful for everyone than becoming a chaos factory of token usage.
Directionally correct metrics are better than no metrics - Token usage might be the best metric in the early days. Especially getting most people out of zero. Getting started does matter, but be thoughtful here because the "for what" matters a lot. It is the job of leadership to and clarify the success outcomes you are focused on, and to make sure you are also measuring progress toward real goals. Odds are early metrics on outcomes are hard. This is why you need to read Patty's book Move - specifically the chapter about control points and limping cows (chapter 4).
Track costs ruthlessly - continue to ask yourself if you are making the right bets and if you are allocating the right amount of capital to the opportunity. Here you must track both AI costs, opportunity costs and human costs. This is not to suggest that you shouldn't be willing to incur extra costs but financial measures are the language of business and it is important for all stakeholders to understand the bets being taken. Cost transparency will do more to drive alignment than you realize. Help bring the organization along with clear financial transparency and measurement.
The pendulum is shifting from no one thinking about this, to people believing that tokens are the new unit of economic value or the new sales and marketing or it's nothing or everything. Winning organizations will make real bets on where and how to leverage AI to deliver (or expand) their strategy. The gap is widening, time to get moving.
Reading and Pondering
Leadership Operating System - I got asked some useful questions on transformational leadership. If you are interested in my take on those topics I recommend you give it a read - thanks Marcin Murawski for the opportunity.
New opportunities of things that weren’t possible before - this is another example of why we must be careful of survivorship bias. The ability to find the new requires you to think past the existing. Not making today better, making tomorrow different (and better). It’s a very different skill and we have to be careful to not focus all our energy in the scarcity story, but make space for the abundance thesis.
On Agents and the enterprise - Ashu Garg gives an update on how agents are progressing both in capability and adoption.
Agent orgs and how alignment goes wrong (Conway's law for agents? Or maybe this is just a new kind of game of telephone ). Meg POV: Systems thinking becomes even more important to best mobilize agents. This is actually just a stronger proof point of why the broder organization and operating model design needs to be top of mind.
Another solid take on the future of HR tech space from George LaRocque - h/t Amy Wilson for flagging and Eynat Guez for helping me to see the bigger picture of the opportunity of tech enablement + deep domain expertise + compliance story that is really about owning an outcome not just a platform.
Things I want vendors to know - an outstanding blog by Alexis Fink about things vendors get wrong in pitching [enterprise] customers.
Shocked to learn that Agents are likely to recommend sponsored content. Useful reminder that there is no neutral, understanding motivations and incentives matter, a lot. When we talk about "judgement and taste" this is one of many reasons why that will matter more in the future and a huge risk as we are all vulnerable to both bias and influence and machines know this better than we do.
Company shape becoming a moat - massively cool post from Jaya Gupta , my thoughts here
A very compelling piece by Tobias Lütke (Shopify) talking about their approach for collaborative learning with humans + agents while protecting the opportunity for on the job capability growth. While I’m here, this podcast was also really fun/inspiring/insightful in the most human + nerdy way possible. And this - don’t overfit for the quantifiable if you prefer something short and insightful . -Meg POV: this is another way to say don’t forget the emotion based data and yes, Meg is becoming a Tobi fangirl. #SorryNotSorry
Deloitte - The state of AI in the Enterprise report - reinforces my prediction for 2026 that AI is moving from all over the place experimentation, to a focused set of objectives. Also that energy is starting to move from optimization to business reimagination – this is all goodness in my world. The curve is going to be quite large for most enterprises as unpacking existing bureaucracy and re-imagining operating models is a heavy lift.
OpenAI and Anthropic create services motions This is interesting to see both from the recognition that the adoption of AI has bottlenecks that humans can help address (as the entire Palantir FDE concept has made clear and also IBM Watson but somehow that never gets the same narrative energy). It is also interesting in the broader context of the growth of foundation model businesses with the recent OpenAI target miss and the broader execution growth of Anthropic. (Anthropic announcement expanding the existing partner network )
Outlier Quotient - great reflection from Adam Shuaib, PhD on how to spot exceptional/generational talent. This tracks both with my future book about the outsiders (specifically the relationship to hardship and the reality that pedigree is often a counter signal). I especially love the reality of discomfort is often the first response and the sense this is related to neurodivergence. Reflecting on my own experiences, with being an overwhelming first impression and I suspect there is some gendered nuance to understand. Both in how neurodivergence shows up differently in women vs. men and the cultural and human reaction to the experience of discomfort from different genders. Lots to think about!
AI Native business transformation example - loved this example as a systems approach to supporting AI transformation in a cross functional way - shout to Claire Vo o for consistently solid content (excited to see her launching additional offerings) and John Kim for the insights.
A good summary of Dorsey mode from Brian Halligan - I can’t help but recognize how much my 2024 Digital Everything perspective has expanded with a stronger unifying activation (hello agents) than I fully appreciated when I wrote it.
Politics of AGI and the potential of “jobless prosperity” - interesting take.
Token budgets running over – between active #Tokenmaxxing and usage monitoring we are finding ourselves in a predictable situation. Uber and ServiceNow are both reporting maxing out budget way ahead of schedule. 20VC chat on the Salesforce token spend math ($300M) Microsoft cancels internal claude code Meg POV: My points last year about both unit economics AND AI accounting are starting to be better understood by the market. Next up will be the re-think of both org budgets and unit economics and this becomes an additional boat anchor for large companies to properly innovate as they have the unfortunate situation of cost scaling at a crazy rate vs. small AI-native org. Also interesting to see people talking about the “end of the token subsidies” which honestly came really quickly vs. past cycles of VC funded tech adoption. My Twitter/X take. #Tokenleveraging > #Tokenmaxxing
Amazon learns tokenmaxxing is expensive (or how math works)
Token pricing and compute
Work and Jobs
Using AI to eliminate your job - You might want to try this. If you leverage AI to make your current job obsolete you are being a high agency person and this path will work well for you -- if you let the business automate your job you will no longer be needed and you will lose your job. This is why so many people are advising you to jump in and learn the tools yourself.
Your job is going away - Lenny's newsletter restates my point above, you need to decide for yourself what happens next. Waiting and hoping is not a good plan.
AI backlash is real, especially for young people - Amy Wilson and I discuss this topic on the pod. Polls show that 70% of Americans think AI is moving too fast, over 50% have negative views of it, and just 18% of young people say they feel hopeful about it.
A useful take on the increased value of domain knowledge - My addition is that domain expertise is great if and only if you are able to use it in a first principles way. If your domain knowledge is exclusively having expertise in existing workflows then I think it’s a liability.
Microsoft 2026 Work trends report - some good nuggets. These are systems problems so require systems thinking. Organizations are now a bottleneck to transformation and leadership is about building new organizations to meet the moment. Owned intelligence is a nice way to think about a specific type of context layer. Thinking of teams and orgs as learning systems is a good approach.
The job situation (openings vs. available workers) - not winning.
Coinbase - the latest “no managers” manifesto. Jason doubles down to say if you are talking about the work your team did you should be fired. Amy’s POV on the “manager wars” (Meg’s add, Adam’s add) and a comic.
Ethnocentrism is a human bug
Enshittification and P(catastrophe)
AI resumes work better for men than women - “When men use AI, we question their effort. When women use AI, we question their integrity. ” (Meg POV: Of course we do)
An inside view of the pre-layoff vibe at Facebook - tl;dr it’s stressful
Regression for women on boards – While women continued to gain seats, men accounted for 77.2% of the change in board composition, 86.2% of seats gained by women were newly added board seats, while only 13.8% came from replacing male directors.
Fun and Funny
Installing a ceiling fan metaphor about Claude task completion bias.
This is old now, but I keep losing it so putting it here so it gets to my Open Brain and for anyone [cough Amy Wilson cough] who might not have heard of the fact that there are now only four jobs. And if you remember only one thing, you must remember that “hot people are the interface layer”.
Lifehacks - AI and Others
While I’ve been starting to get value from my Open Brain project I read Garry Tan’s version and my brain exploded… (for anyone thinking about doing an Open Brain project here are the references from Nate B. Jones and Karpathy (both written for you to give to your favorite vibe coding agent). Claude and I decided to take a few of the ideas from Garry for now and consider the advanced maneuvers for a later date.
Looping Back to You
Thanks for reading, listening, sharing, and helping us think better. We love hearing where you agree, where you disagree, and what you're seeing from your seat.
If someone forwarded this to you, join the Insight Loop to get the next issue directly in your inbox.
And if you're already in the Loop, help us grow: forward this to someone who would enjoy it, and follow/subscribe to The Meg & Amy Show wherever you listen or watch.
Have a question you'd like us to tackle in a Leadership Corner? Send it to: [email protected]
Watch & Listen: Spotify, Apple Podcasts, Youtube, etc.
More from Meg: Visit Meg's Website
** Typos and poor grammar either because I'm too lazy to edit properly, or to make you feel confident this was written by a flawed human and not AI generated.
