Points of Friction
Gil Dibner

The point of maximum change. The VC industry is undergoing another bout of public soul searching and narrative positioning. Perhaps the three best contributions over the past week have been Jack Altman’s hour-long interview with the brilliant Josh Kopelman, a conversation between Rory O'Driscoll and Jason Lemkin hosted by Harry Stebbings (and summarized here by Lemkin), and this keynote talk by Pat Grady of Sequoia. The truth is somewhere in the space defined by these positions. I’ll be revisiting each of these over the next few days as I try to make sense of the environment we are in. That is, in a way, the blessing and the curse of operating (as founders and VCs do) on the edge of innovation. When we are doing it right, we find ourselves caught right at the point of maximum change, uncertainty, and unpredictability - when the world seems to make the least amount of sense. Founders and VCs alike will write and blog and post and speak on stage - all in the effort to convince the world (and themselves) that we have some ability to predict the future. The effort to anticipate what is coming is valuable - but the value lies mostly in the attempt itself, less in the accuracy of the analysis.

The consensus extrapolation on the end of engineering. I don’t think anyone seriously disputes the idea that GenAI has dramatically reduced the cost and complexity of many types of work. This trend is just beginning. Pat Grady framed it very nicely in his talk. What begins as software (a digital tool that humans can use) evolves into a copilot, and then into an autopilot, and then - as Sarah Tavel wrote - into a service that can deliver the work itself. As a software tool moves up the intelligence access, it evolves into a service. This insight applied to the business of software creation itself leads to some pretty damning conclusions for both venture capital and entrepreneurship themselves. Software engineering was originally (and not very long ago) an artisanal craft. Mysterious (and sometimes charmingly anti-social) software engineers (developers! developers! developers!) would sit for hours conjuring up code in what was truly a dark art available only to a very elite few. Around fifteen years ago, the “dev tools” market began - a series of tools (most noteably Github but many many hundreds of others) emerged to provide tooling for these craftsman - whose numbers rapidly expanded as the value of their human efforts were multiplied by these increasingly powerful tools. Today, with tools like Cursor and Github Copilot, we’re witnessing the evolution of the developer tool into the developer copilot. Even more dramatically, we are beginning to see tools like Devin and Lovable and Base44 turn the copilot into an autopilot - largely replacement the work of the engineer. Naturally, there is a growing and not entirely unfounded expectation that soon the ability to deliver the end work product of a software engineering team will become fully productized. At that point - runs the argument - software engineering teams might not be needed in anything close to their current size and scope. There is healthy debate about how far this will go, but I think the consensus view is that it will go pretty far. The days of raising heaps of capital to hire dozens of software engineers may already be over. As the latest data on Series A fundraising confirms - and as Sam Lessin wrote - the days of the “factory model of venture” are definitely over.

Entrepreneurship lives where friction reigns. At the risk of being overly reductionist, the dialogue in technology land seems to be focused right now on how much easier everything is becoming when it comes to building and selling software, and - therefore - on how much faster companies can scale revenues if they can build the right product quickly enough. This is true, and it would be wrong to ignore this (despite a very healthy debate on the durability, signal, and margin of these revenues). At the same time, I am drawn to focus on a different question: in a world of dramatically reduced friction in the production of software, where does friction remain? My conviction is that value is created where friction exists. If I’m right about this, then in order to identify businesses that can create value, I will need to identify companies that operate within some context of difficulty that remains even as the creation of (some types of) code becomes easier than ever. What are the sources of this type of friction? What business plans, go-to-markets, and/or engineering roadmaps are sufficiently hard that winning might afford the victor enough durability to capture real value? I don’t have a Unified Theory of What’s Hard, but I can offer some examples from our recent portfolio:

• Portchain is selling an intelligent software layer into the global ocean freight and terminal industry and is already deployed at over 150 terminals around the world. Maybe the software isn’t rocket science, but selling into that eco-system is pretty close to impossible.

• Sourcix helps industrial procurement teams source custom-machined parts efficiently, saving significant time and cost for some very large manufacturing companies. Their business combines a lot operational heavy lifting to build the supplier network while also deeply integrating into the buy-side, gaining them access to a unique set of training data which simply doesn’t exist anywhere else.

In the midst of today’s zeitgeist of rapidly reducing frictions and mass democratization of the SaaS model, I’ll continue to look for what’s hard to do, hard to build, hard to sell - hard to imagine. If you have a thesis around something that’s difficult and how you’re going to crack it, I’d love to hear from you.

Engineering Success in a Technical Startup
For technical founders, extraordinary success is highly correlated with how obsessive a founding team is about developing an extraordinarily efficient GTM machine.

What if an AI Was Your Best Customer?
Gil explores how AI agents autonomously selecting and consuming tools can reshape the software industry.

Seven Signals
An evolving VC investment decision framework.

Five Emerging VC Swimlanes
The venture market is evolving into five distinct swim lanes, each with its own rules, characteristics, and key players.

ENTERPRISE/TECH NEWS

A2A and MCP. Microsoft CEO Satya Nadella publicly endorsed Google’s Agent2Agent (A2A) and Anthropic's Model Context Protocol (MCP), announcing upcoming support for these open protocols in Microsoft's Copilot Studio and Foundry. (For background, A2A enables standardized communication between AI agents from different providers, while MCP standardizes how AI models securely request external data, promoting interoperability and reducing vendor lock-in.). Rare, and so quite interesting, to see Microsoft to publicly endorse a Google-led innovation.

IDE shopping. OpenAI has reportedly reached an agreement to acquire Windsurf, an AI-assisted coding startup formerly known as Codeium, for approximately $3 billion. This significant transaction, which would be OpenAI's largest acquisition to date, aims to bolster their capabilities in AI-driven code generation and developer tools. The move positions OpenAI to strengthen its competitiveness in the rapidly evolving market for AI coding assistants, challenging existing players like GitHub Copilot.

HOW TO STARTUP

When AI CEOs? Benn Stancil put out another great piece in his newsletter this past week. In it, he argues that AI is not merely a new feature but is fundamentally integrating into and transforming all sectors of technology and industry. The crux of his argument is as follows. LLMs particularly good at synthesizing and summarizing lots of disparate content. Our vision for most AI-first tools is to use them to transcribe and store vast amounts of information (e.g. Gong or Granola to transcribe every sales call with the promise of making them searchable/understandable). In a world where everything is transcribed, however, are we humans best positioned to synthesize all that infinite content? Or are the models themselves?

YCs RFS. Always interesting to read YCs requests for startups. Take a read here.

Augment or replace? Fiverr CEO Micha Kaufman sent an email to employees stating that AI poses a significant threat to many jobs, including their own and his, calling it an "unpleasant truth" and a "wake-up call." He stressed that employees must quickly adapt and become adept at using AI tools relevant to their fields to remain competitive and avoid their skills becoming obsolete in the rapidly evolving job market.

HOW TO VENTURE

Coatue’s new fund. Coatue Management is launching a new investment fund, CTEK, focused on AI and enterprise software. Uniquely for the firm, this fund is open to individual investors with a significantly lower minimum investment of $50,000, a substantial drop from their usual $5 million requirement. (How much of that $50K will be eaten up by fees?) CTEK is starting with $1 billion in initial backing from the family offices of Jeff Bezos and Michael Dell and plans to offer investors liquidity through quarterly tender offers.

Venture barbell. A great summary of the current state of affairs by Samir Kaji. The industry has evolved into a "barbell" structure, Kaji argues, no longer representing a single asset class. This structure comprises smaller, early-stage funds seeking outlier returns with high dispersion and large, multi-stage platforms operating more like private equity with diversified strategies. The author contends that Limited Partners (LPs) must recognize this divergence and apply different underwriting and allocation strategies tailored to each distinct side of the venture capital barbell.

Thrive’s big bet. Anysphere, creator of the popular AI coding assistant Cursor—which lets developers use natural language to generate code—raised $900 million at a valuation of roughly $9 billion, more than tripling its value since January. Investors including Thrive Capital, Andreessen Horowitz, and Accel are betting heavily on the rapid revenue growth (reaching $200 million annual recurring revenue by April) as companies increasingly adopt AI tools to boost programmer productivity. This deal reflects a broader surge of investment in generative AI startups, despite lingering questions around the sustainability of high valuations amid volatile public markets.

FalkorDB was selected to join Batch #17 of SV101 by ICON.

Groundcover’s recent $35M Series B to take on legacy observability vendors was covered in USA Today.

Aquant
Sales Account Executive - AI (Boston)

Firebolt
Director of Product Management (Bay Area or Seattle)

Portchain
Support and Data Specialist (Copenhagen)

Sourcix
Head of Marketing (Tel Aviv)