Thoughts on the Software/AI Selloff

[Smorgasbord1]

Coming back to DataDog, I found this video interview with Lemonade’s CEO Shai Wininger talking about his company’s use of AI:

Lemonade CEO Shai Wininger on How Lemonade Rebuilt Its Product, Org, and Stack for the AI Era

It’s a half-hour, but here are pointers to two sections on his proposal to replace DataDog with AI:

12:49 - A set of agents capturing issues (it takes him a minute to get to this, so be patient, or jump to 13:50)
“We’re talking about…system reliability, system diagnostics, and observability. The next generation would be not having those things at all. If you have the right sensor running in production and it captures that not only that there was a bug, but the source of that bug, then you have the right context for an agent to do the fix and test it…without having a human…A self-healing system.”

16:00 - Mentions DataDog specifically. “Right now we’re using the standard observability platforms like HUD and DataDog to first collect all the anomalies, and then qualify them for potential fixes.” Unfortunately, before he can complete that thought, he distracts himself with a “funny story.”

Lemonade is an AI native company, and has been for years. If Wininger is talking about replacing DataDog now, other companies will be thinking about it in a year or so. And the LLM companies will themselves want to show what their products can do, so look for Anthropic plug-ins or OpenAi skills that start to do the detection, and add isolation and eventually proposed bug fixes. Might take a couple years, but I do firmly believe it’s gonna happen.

BTW, at 27:49, Wininger talks about how Lemonade’s user interface is almost 100% AI chat (Maya and Jim).

[intjudo]

So the logical extreme is

1. AI enables us to do things right in the first place
2. Result: dramatic reduction in the amount of data generated
3. Result: lots less for AI to process!
4. Result: AI compute computes itself out of a job

Are we going to see a increase in the demand for AI compute because AI is great?

OR are we going to see a dramatic reduction in the demand for AI compute because AI is so great?

[Smorgasbord1]

@intjudo

AI compute computes itself out of a job

Echos of 1940’s Thomas Watson’s (CEO of IBM at the time) statement that he thought the world market for computers was about 5 are ringing through my head.

Look up Jevon’s Paradox.

The cheaper and better AI gets, the more it will be applied. We’ve barely started scratching the surface. The balance of where compute lives will change (from data centers to edge devices like cars and robots), but data center compute is far from topping out.

[intjudo]

@Smorgasbord1

The cheaper and better AI gets, the more it will be applied.

Well yes; but also and perhaps more importantly:

The cheaper and better that the systems AI is applied to (“AI beneficiaries”) get, the more demand for the AI beneficiaries will increase.

So if AI makes processing of streaming data, analysis of log files etc. more cost effective (…EVEN IF it does so by drastically reducing the volume of data that needs to be processed in the first place), does that mean we need LESS processing of streaming data and log files?

I think maybe it means we’ll have an increase in the TOTAL demand for processing streaming data and log files. If the AI beneficiaries are less expensive to deliver because costs involved with streaming data and log files go down, it should result in more IT stuff getting built, because the demand for IT stuff is infinite.

Paradoxically, an external force that reduces volume for datadog to process could end up creating additional demand for datadog’s services. After some short term pain perhaps. How short? Who knows

[Smorgasbord1]

Back to another AI/Software stock - anyone here still in Lemonade (LMND)?

They announced earnings a few days ago, and the report looked good: Revenue up 53% to $228m, gross profit up 73% to $111m, $73m of free cash flow, but still a net loss (although smaller loss from $0.42 to $0.29 per share)

Guidance was up, too.

But, I think the valuation is too high. Not profitable, so looking at price to sales, the ratio is over 8 while the non-AI insurance companies are like 5 or under.

I bring this up not so much for potentially investing in LMND (although that’s not out of the question for me), but to point out how inconsistent the market is: Take down legacy names because AI will disrupt them and then take down an AI-native company working on disrupting the legacy providers as well.

[FoolishJeff]

I have a small position in LMND, it is definitely not cheap but looks pretty reasonable now with the recent haircut. NTM EV/Revenues is 3.1 vs PGR whose NTM EV/Revenues is 1.43. Obviously, LMND is growing much faster. Also, LMND is on track to be be profitable this year.

[SlowAndFast]

I guess we are getting close to having beaten this topic to death, but heard this interesting CNBC interview with Nvidia’s Jensen Huang today, where he essentially makes the following points:

Nvidia CEO Jensen Huang on AI's pressure on software stocks

• AI Agents can write software.
• The tools companies / platform companies / independent software vendors / SaaS companies, will use agentic systems to develop their software.
• But enterprises will not develop software, they will use Agents to use those tools.
• So agents will not replace the tools, but will use the tools to help us be more productive.
• And because of the explosion of agents out there, the tool use will actually go up (and not down).
• Another point he made is that the software companies’ agents have to be experts in what they do and the software companies already working in this space will have the necessary edge (and training data) to fine-tune those agents.
• He ends with an optimistic view that instead of replacing us humans, AI agents will help us: The purpose of the work that’s going to be replaced by agents won’t change. So for example, a software engineer won’t need to code anymore, but instead describe to an AI agent what their intentions are and what kind of ideas they have and the agents will write the actual software. So we’ll be working at a different level of abstraction and be much more productive as a result.

Ben

[Smorgasbord1]

It’s probably worth defining what an “AI Agent” is, since the term is bandied about a lot and I suspect many people don’t understand it.

An AI Agent is a program that actually does something. For instance, classic ChatGPT is NOT an agent - all it does is talk back to you.

You’ve probably heard of “Open Claw” - that’s a framework in which you can create/deploy AI Agents to do things like organize your email, send text messages, or make restaurant reservations for you. Sometimes, but not exclusively, an AI Agent is not itself an LLM, but rather uses an LLM (this is how OpenClaw works). Claude Code is an AI Agent that I believe has its own internal LLM that it uses to write code, as well editing files, running commands, checking in files to Github, etc.

Huang’s view that AI Agents won’t replace the software tools being used by humans is reasonable, but maybe short-lived. This gets us back to the original discussion here about AI Agents replacing human seats in SaaS. It’s possible for customers to deploy AI Agents to run SaaS like Workday or Salesforce or even DataDog, and since those AI Agents work fast and around the clock, customers deploying them may not need to purchase as many seats of those SaaS products. And, this is done by creating AI Agents, which is not the same complexity as creating the LLM AI itself. OpenClaw, for instance, was created in a short span of time by one person. All “vibe coded,” btw. So, yeah, it’s probable that AI will just use the existing SaaS tools, if the SaaS vendors keep their functionality strong and unique. But, there is always the possibility - probable in many instances I feel - that AI can do a better job at many functions for which SaaS tools are used today. And some SaaS companies are adding those capabilities, to be fair.

But, I feel Huang is wrong about not replacing humans. Not only will a software engineer not need to code, but since he only needs to describe to the AI (called “prompting”) what is needed, then you may not need that engineer at all. A detail-oriented Product Manager could feed the AI his product spec (what it needs to do) and get something workabout out of it. And since non-AI coded projects take many engineers, overall the human staff of software engineers will be reduced.

And we just got confirmation on that today with Jack Dorsey’s company Block (ticker: XYZ) announcing a massive layoff (40% of the company = 4000 people) due to AI. Block CFO Amrita Ahuja said: “We see an opportunity to move faster with smaller, highly talented teams using AI to automate more work.”

[Smorgasbord1]

Snowflake announced good earnings, and so its stock went down. This is today’s market.

But, rather than talk about results, I wanted to bring up a couple items from the ER call, which was very interesting, at least to me:

1. Snowflake is partnering with SAP. Not Salesforce.com, but old-fashioned, not yet dead despite being disrupted by the cloud, SAP.

2. Snowflake completed its acquisition of Observability company “Observe.”

3. Snowflake has its own AI tools builtin (Cortex Code and Snowflake Intelligence). They’ve eaten their own dogfood, apparently on two fronts:

“Across our business, Snowflake Intelligence and Cortex Code are already delivering measurable results. Our service delivery team can complete customer projects up to five times faster, improving response accuracy by more than 25%, and compress implementation cycles from days to hours to drive 40% to 50% higher project margins and enabling customers to go live more than 40% faster.”

and

“We have seen our site reliability engineering issues that once required hours across multiple engineers now resolved in minutes, dramatically reducing resolution time and further strengthening Snowflake Inc.'s reliability. And we have built agent capabilities that help our sellers prioritize accounts, automate research, and generate personalized outreach, projected to recoup the equivalent of 90 full-time engineers of productivity this year. Our finance team is working on automating travel and expenses analysis, proactively curbing out-of-policy behavior—an initiative that is expected to drive millions in annual savings.”

This sounds to me like Snowflake is replacing their previous CRM system (Salesforce.com?) with their own AI-based and AI-coded product, and also already switching to Observe with AI versus whatever they were using before. These are saving the company millions of dollars and improving company response time/performance.

Now, Snowflake is obviously highly technical, but I suspect most of this was done via “Vibe coding,” and so Snowflake reps could demonstrate to their customers how they could copy Snowflake’s approach.

And, back to one of my original posts in this thread, this is why OpenAi as a probable customer of DataDog might be looking to do something on its own to both save millions as well as have a product to sell to its customers.

Disruption is here, but I do admit that these changes won’t diffuse into the world all at once, but over years. Regulated industries, government, very large organizations will obviously take longer, but there’s low hanging fruit as well. And new companies are being started today to work from scratch as AI native.

[wpr101]

There has been some really strong confirmation from both the hardware side and software side that Q4 saw agentic AI take off in a massive way,

Cloudflare NET said this about agentic AI,

Over the month of January alone, the number of weekly requests generated by AI agents more than doubled across the Cloudflare network. This is driving increased demand for our whole platform. This is where Cloudflare’s scale becomes our moat. With more than 20% of the web already sitting behind Cloudflare’s network, we are effectively the global control plane for the agentic Internet. That’s creating a number of new growth opportunities, both with our traditional business as well as what we’ve begun calling Act 4, helping invent the future business model of the Internet.

A couple quotes from Nvidia as well,

Frontier agentic systems have reached an inflection point. Claude Code, Claude Cowork and OpenAI Codex have achieved useful intelligence. Adoption is skyrocketing and tokens are profitable, driving extreme urgency to scale up compute.

Anthropic’s Claude Cowork agent platform is revolutionary and has opened up floodgates for enterprise AI adoption. Between Claude Cowork and OpenClaw, Anthropic’s Claude Cowork agent platform compute demand is skyrocketing and ChatGPT moment of agentic AI has arrived.

We have now seen the inflection of agentic AI and the usefulness of agents across the world and enterprises everywhere. You’re seeing incredible compute demand because of it. In this new world of AI, compute is revenues.

---

It seems like there will be a lot of implications for the companies the board follows for how they can benefit (or be hurt) by agentic AI.

These statements above also seem to squash the notion that spending is going to suddenly stop, especially if the companies are getting a return on investment and that “compute is revenues”.