What’s 🔥 in Enterprise IT/VC #479
Wait Till Agents Really Work: Fortune 500 Margin Expansion Before the AI Wave Hits
🤯 This comment caught my eye and I’m sure most of this efficiency is still from the Post-ZIRP hangover of insane inefficient hiring. Many of these F500 CEOs have talked about AI as driving this but frankly I’d say that most of that is more marketing than reality.
That being said, this is simply insane efficiency gains before AI has really proliferated. The budgets of these top 100 companies, many leading banks and tech companies, are massive and the appetite for automation is even larger.
Claude has already been a killer productivity boost in 2025 - here’s a principal engineer at Google sharing her thoughts.
Seems like in 2026, there will be some insane super powers beyond just coding 🤯 - 2026 will be a huge year for AI as agentic automation.
But Manus (acquired by Meta for over $2B) seems to have already been on this by optimizing for solving problems versus benchmarks
Also ties in really well with what’s needed to make it even better which I posted in last week’s newsletter on the Execution Intelligence Layer for Agents.
In Satya Nadella’s year-end reflection he muses on what’s needed - rich scaffolds that orchestrate multiple models and agents; account for memory and entitlements;…(it’s short and read his musings here)
As far as who wins, lots of startups will go after this but it is so hard to be a middleware layer. I assume the biggest winners will just build this into their own stacks from incumbent SaaS providers like ServiceNow and Salesforce and the larger LLM providers as well. To win, IMO, startups will have to have incredible time to value with a narrow, vertical use case with complete out of the box integrations and intelligent context in areas like finance, IT, and security. Of course, I do have one of these in the portfolio and can’t wait to share!
So agree with this from Animesh who wrote a follow up article to Jaya on How to Build a Context Graph. Point is we need a next-gen Celonis or FortressIQ, that is intelligent and learns and maps how people interact, what data flows, and map out workflows.
Key IMO will be not only mappying those out but also perhaps running simulations to improve them. There are so many processes in organizations that are just there because that’s the way they’ve been done for decades. Turning these relationships and processes into data with today’s AI will allow you to make these even better. Why automate something that already exists if you can squeeze more efficiency from doing them differently? You just need the result to be consistent in the end.
Barron’s is on this for 2026!
Happy New Year to all and look forward to seeing you in 2026!
As always, 🙏🏼 for reading and please share with your friends and colleagues!
Scaling Startups
#Peter Walker has the definitive data set for startups for 2025 - a few interesting charts for those founders going from Inception/Seed to A - this is median and as FYI many of my port cos had minimal ARR and raised outsized A rounds
#easier said than done - my prediction for 2026 sounds obvious but much harder to execute - also click through for the full recap of what my fund boldstart ventures was up to in 2025!
Going to be an epic 2026 as agentic workflows move beyond just coding to deliver more value in the enterprise...but lots to build 🏗️ and a word of caution
#so glad this mode is back - i just call it #winning
#want more enterprise design partners👇🏻 - some tried and true methods I’ve learned over many investments and decades 👴🏻
#💯
Enterprise Tech
#Manus - Meta 🤯 - congrats to Benchmark on investing in this team from China who moved to Singapore and also signals that agents without a research lab can be incredibly valuable to acquirers 🤔
Brian from Sequoia chimes in
Greg Isenberg shares a great take👇🏻 for comments
#one of the themes of 2026
there it is again - “true authenticity becomes scarce”
#i highly recommend checking out the Google Deepmind job board to understand priorities for 2026 - 93 open roles give you a glimpse into future
#selling AI workers - btw, my friends have been doing this for quite some time with some insane success
Dave Elkington and Gabe Larsen from Signals
#the year ahead in robotics from Nvidia’s head of robotics - software is the limiting factor!
#learn how to use Claude Code from the creator
#don’t let super volatile token prices fool you, it’s going to be a big year for crypto infra and TradFi in 2026, lots of tokenization of assets and stable coins and more
#spot on
Markets
#in 2026, beware of China, government backed VC funds for AI and hard tech, IPOs of AI companies giving them lots of cash, and insane cutthroat competitive dynamics - must read letter 👇🏻
#if you’re over 40 and don’t understand why betting markets and the rise of casino culture, i highly recommend reading this 👇🏻
#lots of questions around the Nvidia Groq $20B license/acquihire on who gets paid - Dan Primack from Axios did some digging and the answer is “everyone gets paid alot!”
