2025-05-08 08:00:00
With models rapidly commoditizing in performance, we are seeing different strategies for keeping users on the same models.1
Ultimately, personalization & applications are likely to be the two vectors by which foundational model companies compete.
A model remembering your name, optimized to your codebase, having a knowledge of your previous work & refashioning itself to the way you work : those are reasons for loyalty, even if the model isn’t state of the art. Like airplane reward programs, personalization & memory introduce switching costs that may outweigh the benefits of state-of-the-art models.
Fidji’s hiring & role suggest Applications for OpenAI are of paramount importance.
The title CEO of Applications implies Fidji will be the head of a substantial & strategic business unit & OpenAI will evolve to a holding company structure - like Sundar’s Alphabet with a CEO of Waymo (actually co-CEOS!) & YouTube (Neal Mohan) for example.
The personalization of the model within an application & workflow will be the ultimate moat. OpenAI’s recent leadership hire makes plain how large the opportunity is.
1 A recent paper on AI competition suggests that newer models are rapidly adopted by customers. Sometimes newer models sap share from competitors, other times replace previous versions (Claude 3.5 users upgraded to 3.7), last new models expand usage (Gemini Flash & 2.5).
2025-05-07 08:00:00
As traffic to traditional websites plummets due to AI answering user queries directly, there is a new explosive form of distribution.
“AI agents are now creating Neon databases at 4x the rate of human developers, driving new requirements for instant provisioning, automatic scaling, & isolated environments.”
If I ask an AI agent to create a web application, I want it to select all the components : the front end framework, the database, & the hosting service. I just want the website to work.
In that modus operandi, the person (the vibe coder) is now no longer involved in vendor selection.
The agent won’t pause to ask, “Do you have a preferred vendor?” busy reasoning about the task at hand, it will continue on its code path, hurtling towards an answer.
AI becomes the ultimate influencer, more than a social media magnate, or an engineering luminary, or a starry open-source repo - an outsourced procurement team acting on behalf of the coder.
This example makes clear how developer tools will be selected by AI. But in a future where sales development reps are assembling data pipelines with data enrichment or email providers or website builders for account-based marketing, passive AI procurement will move up the stack.
Product-led growth companies have a new & important channel to master : AI as the Chief Procurement Officer.
2025-05-05 08:00:00
Election night struck the regulatory asterisk from web3. But it did more than that.
It triggered a broader shift of application investing versus infrastructure.
The last few years of crypto & Web3 investing have focused predominantly on infrastructure, the databases (called Layer 1s/L1s & Layer 2s/L2s), security, analytics, & decentralized finance or DeFi (typically lending products). But these categories are slowing.
In 2025, gaming, real-world assets (RWA), payments, & applications have all captured 10% more venture dollars compared to pre-election.1
The magnitude of these changes is not huge, a +/- 10% statistically significant change across about $1.6b in disclosed investments since Election Day but the data portends a broader shift in Web3 investment activity.
First, stablecoins are rapidly becoming a new & valuable payment pathway for large corporates with Stripe, Visa, PayPal & many others supporting these rails. New banks, new payment processors, & merchant acquirers have emerged to meet the market opportunity, building payment transfer applications & accounting businesses on top of these new corridors.
Second, the infrastructure advances of the last few years have improved the price/performance of Web3 blockchains, enabled & improved user onboarding for games, both through better wallets & also developer prepayment of fees on behalf of users.
Third, some Web2 financial instruments are moving to Web3 & bringing with them superior benefits. Daily dividends, instant settlement, 24 hours a day, & also programmable collateral to drive more yield than is available elsewhere.
This evolution is a wonderful thing.
For many quarters the dominant theme at crypto events has been “wen application?” When will the first killer applications on Web3 be built? If venture capital behavior is a credible predictor of the future, they are already here.
They may not all be trumpeting the fact that a Web3 database is powering them under the hood. But that’s the point. The underlying technology is an enabler to a superior end user experience. No asterisk needed.
1 This change is statistically significant. The Chi-squared test resulted in a p-value of 0.0356 & a Chi-squared value of 27.6.
2025-05-01 08:00:00
“We processed over 100t tokens this quarter, up 5x year over year, including a record 50t tokens last month alone.”
If the market harbored any doubt for the insatiable demand for AI, this statement during Microsoft’s quarterly earnings yesterday, quashed it.
What could this mean for a run rate? Using some basic assumptions1, this implies :
| Scenario | Model mix (% of total tokens) | Monthly run-rate after 20 % discount | Annual run rate | % of Azure Revenue (assuming $21B Annual) |
|---|---|---|---|---|
| High | OpenAI 70 % • Claude 20 % • Other 10 % | 382.9 | 4,594.8 | 21.88% |
| Medium | OpenAI 65 % • Claude 20 % • Other 15 % | 110.5 | 1,326.0 | 6.31% |
| Low | OpenAI 60 % • Claude 20 % • Other 20 % | 27.3 | 327.6 | 1.56% |
So AI is roughly between 2 to 22% of Azure revenue. Error bars here are quite large, though.
A major contributor to this increased demand is performance, especially with reasoning models.
Combined with some of the massive reductions in inference costs, especially with smaller models like the Phi-4 models that Microsoft released yesterday that are open source and small. The margins on AI inference should continue to surge.
“…our cost per token, which has more than halved.”
“You see this in our supply chain where we have reduced dock to lead times for new GPUs by nearly 20% across our blended fleet where we have increased AI performance by nearly 30% ISO power…”
Jevon’s Paradox in full force.
“The real outperformance in Azure this quarter was in our non AI business.”
This was a surprise, but it likely is the result of additional demands placed on adjacent systems. AI doesn’t exist in a vacuum. It needs databases, storage, orchestration, and observability to succeed.
“PostgreSQL usage accelerated for the third consecutive quarter… Cosmos DB revenue growth also accelerated again this quarter…”
A later quote within the analyst call reinforces this point, the database systems, Cosmos (a MongoDB-like document data store) & PostGres, Both of which are transactional databases.
100 trillion tokens up 4x y/y. Next year, could we see a quadrillion?
1 20:1 input-to-output token ratio; a model usage mix of 60-70% OpenAI, 20% Anthropic, remainder of other models ; and a 20% discount to public prices. See the work here
2025-04-30 08:00:00
By 2026, AI agents will consume 10x more enterprise data than humans, but with none of the contextual understanding that prevents catastrophic misinterpretations.
In this presentation I shared yesterday, this is the main argument.
Historically, our data pipelines have served people. We’ve architected complex pipelines to ingest, filter, and transform information in different systems of record: cloud data warehouses, security information and event management systems (SIEMs), and observability platforms.
We then interpreted these outputs and acted upon them.
But very quickly the end consumer won’t be people. So, we need to fundamentally reconsider the interface between these systems of record and their transformed data.
People thrive in ambiguity because we’re great at contextual interpretation. One VP of Sales mentions revenue, a CFO understands the demarcation between bookings, billings, GAAP revenue, or contracted ARR. Humans navigate these nuances effortlessly, machines don’t.
What happens when your AI agent pulls “customer acquisition cost” data but doesn’t recognize that marketing measures it by campaign spend, sales calculates it based on AE + BDR costs, & finance includes fully-loaded employee costs?
The result: expensive nonsense masquerading as intelligence.
To combat this disinformation, the teams that were formerly responsible for maintaining and monitoring pipelines will become cultivators of a constantly evolving collection of cross-domain semantic layers that feed the questions from AI agents via MCP or another protocol layer.
The major question in all this is how to deliver the semantic layer. Historically, it’s been difficult to sell a semantic layer as a standalone product. Looker was successful with its LookML language, and other companies have developed their own query language, which to some extent has enforced a loose semantic layer.
The coming years will see a major shift as enterprises realize that their most valuable digital asset isn’t their data lake or their AI models—it’s the semantic layer that makes those investments meaningful.
Software is the business of selling promotions, and no one has been promoted for implementing a semantic layer. However, many people will be promoted for massively improving the accuracy of AI systems and across data security and observability.
The semantic layer is the keystone to that project and consequently, the most strategic part of any data pipeline today.
2025-04-28 08:00:00
If every employee starts managing agents, how does a company change?
First, “83% of global leaders say AI will let employees take on more complex, strategic work earlier in their careers.”
One executive recently framed this transition of teams evolving to three areas of work : operational, tactical, & strategic. Operational work can be mostly fully automated today. Agents are chomping away at tactical work - better accuracy will improve their share. Humans will focus on strategy work, but likely assisted by AI.
“When asked why they turned to AI instead of a colleague, employees cited 24/7 availability (42%), machine speed and quality (30%), and unlimited ideas on demand (28%) as the top reasons—all things humans cannot provide.”
Advanced AI users use AI as a thought partner. With some studies demonstrating AI’s creativity as superior to humans, combined with infinite patience, profound memories, & a little bit of obsequiousness, we should expect AI as a strategic work sidekick.
The impact on org charts is likely to be profound.
Manu Cornet’s interpretation of org charts will need an additional box. What will it look like?
“But with expertise on demand, the traditional org chart may be replaced by a Work Chart—a dynamic, outcome-driven model where teams form around goals, not functions…”
Microsoft’s Work Trend Index argues most businesses will resemble movie production : teams of specialists who descend upon a project, achieve a goal & move on.
And the impact is mostly on customer-facing teams : “In our survey, global leaders listed customer service, marketing, and product development as the top three areas for accelerated AI investment in the next 12–18 months.”
The Cornet image of the future will undoubtedly have more agents than humans - by 10x or 100x is hard to say. The leverage from AI is hard to overstate & expectations for speed, depth of thought, creativity, & effectiveness will surge as a result - a huge opportunity for those who understand how to use these new tools effectively.
