Загрузка видео...
Не удалось загрузить видео
Chamath: Software IP Just Became Worthless “I think that IP and software was effectively rendered worthless two days ago. There have been people that decompiled the entire Big 5. What is the Big 5? Adobe has five very, very important products in their product suite, and these folks basically... show more
304,903 просмотров • 2 дней назад •via X (Twitter)
Комментарии: 33

@chamath Fun fact: moats weren’t to make it hard to cross, they were to make the ground impossible to dig under. Before moats people would bring walls down by digging under them until they collapse.

@chamath This all makes the assumption nothing new needs to be created and that everything great already exists.

What is the actual number of customers who will drop Adobe to use open source? In the big scheme of things probably not that many. I know I won't. We were told for decades that Linux was going to destroy Microsoft and it never went anywhere of any substance. If anything it will force existing companies to step up their game.

@chamath I think guys like chamath are too smart to sound that stupid. So I assume it’s just an affectation / deceitful way of communicating.

@chamath

@chamath It's actually funny that Silicon Valley pushed Ai down the throats of ordinary citizens so hard that now Ai is going to bankrupt the whole ind We used to be able to buy Office once and never have to again. Silicon Valley is original Sony in terms of you not owning anything

@chamath I don't have to find the 'cracked' version that won't install malware on my machine anymore?

@chamath Well the product still needs to work as good as the original one.

@chamath Chamath declaring software IP worthless is rich coming from a guy who monetizes his opinions.

@chamath That IP was ever something forever was a joke of their own making. All inventions should have a shelf life of no more than 5 years. That would encourage continuous inventiveness.

@chamath Yep. Massive enterprises are all going to flock to use some random tools that are now available first thing Monday morning. Don’t even ask their IT, security or legal teams. Just move everyone right over and it’ll all work out perfectly. Up and running by EOD.

@chamath i don't buy worthless, a decompiled copy still needs somebody to maintain it

@chamath Hey this was us. We'd love to chat @Jason @chamath

@chamath If the code can be decompiled, the code was never the moat. The moat was knowing what to build and who to build it for. A decompiled product isn't the same product. It's the same features with nobody deciding which ones matter next.

@chamath Chamath is confused. They didn't decompile it. That would be illegal and it would have been taken down by Github. They did a clean-room recreation, which is the legal version.

@chamath I think your framing as 'decompiled' is wrong. But the larger implication stands. They claim, wrote from scratch using Rust and published on Github using permissive open source licenses.

@chamath Would love to do a chat with you guys. @chamath @Jason @DavidSacks @friedberg - David is especially right about open source and open weights.

@chamath Yeah the big questions looms ! Why can’t this be done to system design , hardware design , chip design and eventually rack each of them as mcp servers and tell your agent , now go do your thing !

@chamath I’m glad I pivoted to services and away from core software development, I agree the IP and moats in software have all but dissapeared. I’d rather optimize an existing services business with Si and automation, I’ll gladly take market share and healthier margins along the way.

@chamath So they used AI to steal intellectual property and repackage it. Are AI agents becoming the next China?

@chamath Sftware IP isn’t dead. It’s just having identity crisis. When open-source alternatives can replicate years of developmnt, the real moat may no longer be the code, but the data, distribution, trust, and execution around it.

Maybe you guys should teach Friedberg the difference between "silicone" and "silicon"... He confused the two multiple times during the episode confirming it wasn't a mistake, per se. He just doesn't know the difference. Anyone so involved in the tech industry should probably be aware they're not the same... I'd hate for him to spend all his time developing software with fake tits that's only good for sliding up and down a brass pole.

@chamath Building Adobe with Claude is like riding a jet ski from LA to Honolulu. Plausible. Practical? No.

@chamath The west needs to decide now if distilling is copyright stealing. Because that's what it is tbh

This has been a problem for people building ai solutions that solve ai problems for at least a year. This is just the clearest and biggest (so far) example of how IP law doesn’t work in the ai era because it shows that traditional software is the emperor who has no clothes.

@chamath The code was never the moat. Decompile Salesforce tomorrow and you still do not have the fifteen years of records inside it, the integrations writing to it, or the argument everyone already had about what a closed deal means. That is the thing companies are actually locked into.

This is because real security was never put into place for computers. If there was real security, all code would be encrypted going in and out of the CPU. The cost would have been negligible and essentially no impact on speed. Data can be encrypted to every CPU, GPU, MCU, but companies were lazy. Now everything is insecure.

Chamath’s claim is right for Software TOOLS and Apps like Adobe. It is not correct generally. Ecosystems, platforms and enterprise layers like Service $NOW are in a fundamentally different position. Here some of my discussion with Grok about it: Enterprise buyers already treat ServiceNow (and similar platforms) as a system of record and system of action. Adding AI agents does not remove the need for architectural integrity, permissions, auditability, compliance mapping (NIST, EU AI Act, etc.), data residency controls, and a reliable kill switch. Replacing that layer wholesale with a collection of agents and frontier models introduces material operational and regulatory risk that most large organizations will not accept quickly. The switching costs and blast radius of failure are simply higher than with a standalone app. ServiceNow is actively leaning into the “use us for AI” path rather than treating agents as pure competition: • AI Control Tower has expanded into a five-pillar control plane (Discover, Observe, Govern, Secure, Measure). It inventories agents, models, prompts, and datasets across ServiceNow and external systems (AWS, Azure, GCP, SAP, Oracle, Workday, etc.). It includes runtime monitoring, least-privilege identity governance (via the Veza acquisition), and real-time containment/kill switches for agents that go outside their scope. • Action Fabric + MCP Server deliberately opens governed, headless access so external agents (Claude, Copilot, custom stacks) can invoke ServiceNow workflows and data under the platform’s existing permissions and audit trail. The goal is to become the trusted execution and control layer that agents connect to. • Multi-model, model-agnostic approach. Leadership (including Amit Zavery) has stated that frontier models currently have limited durable moat and function more like interchangeable infrastructure. ServiceNow routes work across partners (OpenAI, Anthropic, its own smaller open models such as the Apriel family, etc.) while keeping governance, context, and customer data inside its perimeter. Less than 10% of the platform’s IP is described as the LLM itself; the rest is process context, workflows, and controls. Your point about commoditizing frontier models while supplying the secure, cost-efficient access layer is consistent with this direction. If models continue to converge on capability while price/performance improves and open or smaller specialized models close the gap, the scarce assets become reliable orchestration, identity, audit, and deep process context. ServiceNow is trying to own that layer and let customers (or their agents) pick models more freely underneath it. Caveats remain. Interface ownership can still shift if agents or other surfaces become the primary user habit. Pricing is already moving from seats toward consumption/outcomes. And over a longer horizon, better agent frameworks plus lower migration costs could reduce some of the lock-in. But for the majority of regulated or complex enterprises, the near-to-medium-term rational path is indeed to put AI on top of or through the existing platform rather than attempt a full replacement. The risk/reward of ripping out the governance and execution substrate does not currently favor replacement for most of them.

@chamath Tell that to the crowd — we score 22 finance channels and NOW is 79% bullish (96/25/1) across 122 videos at $140.86 vs an $81–$193 52w range. If software IP is worthless, why do 96 videos still say buy? Is NOW the exception or the illusion?

@chamath @Grok true?

@chamath It's not "decompiled", but AI rewrite. They are fundamentally different.

@chamath AI把IP干成白菜价之后,软件股的护城河是不是只剩分发了?

@chamath 🎯💯
