About
Ed Sim is the founder of boldstart ventures, a $1.1B AUM venture firm that invests at…
Articles by Ed
Activity
18K followers
-
Ed Sim posted thisis the SaaSpocalypse truly over? Marc Benioff certainly thinks so, and Salesforce shared some some numbers to back it up: → 3.2B agentic work units in Q2, up 97% q/q → 6x growth in MCP usage → top AI companies expanded their Salesforce spend 435% YoY 🤯 → seats are still growing enterprises still need systems of record for permissions, business rules, audit trails and governance EVEN in a headless world Read more in What's 🔥 in AI/Infra/VC #513 (full post in comments)
-
Ed Sim reposted thisEd Sim reposted thisOver 100 companies signed an open letter today saying we have a limited window on cyber defense. Snyk signed it. The most important word in that letter is "verified." Nothing in it says who does the verifying. That question is the whole game now. AI made finding vulnerabilities close to free. It made trusting your software expensive. If the model that writes the fix also certifies the fix, we have not reduced risk. We have industrialized unverified change at machine speed. The generator cannot be the validator. That is not a vendor position, it is separation of duties applied to machines, and it applies to the agents we build too. Every CISO I talk to is asking a version of the same question: how do I know the fix is real? That is what we are building into the Snyk platform. ✅ Deterministic validation, independent of whatever wrote the code. ✅ Frontier reasoning grounded in Snyk Security Intelligence, so the model reasons over confirmed findings instead of plausible ones. ✅ The Remediation Agent applies the fix, rescans, runs your existing tests, then hands your engineer the PR. A human still signs off. The output is a fix that survived a check, not a suggestion that joins the backlog. Labelbox cleared a two-year backlog in weeks running this workflow. Now every Snyk customer can run the same loop. And the internet's first line of defense is still unpaid open source maintainers. They should not be doing this with consumer-grade tools. Through our Secure Developer Program, qualifying maintainers get a Snyk account with full enterprise entitlements and no usage limits. The same product our largest customers run. Apply: https://lnkd.in/eNkm2rtJ Signing a letter is the easy part. Put capable defense in more hands, then measure progress by issues fixed, not issues found. OpenAI Snyk
-
Ed Sim shared thisAs you know I'm obsessed with digital agents. Chewbarka, my own OpenClaw bot, now GrokBot. That's real, and it's happening. What we forget is that most of the world's work still happens with hands, not keyboards. Half the planet is still in farms and factories. Everyone else is in warehouses, hospitals, restaurants, trucks. The holy grail IMO is physical. The dream state is a robot that watches you do a thing once and then just does it. Until now that hasn't been possible. This week Generalist (a boldstart ventures port co) launched GEN-1.5, a one-shot learner. Robot see, robot do. No longer a chat prompt. A Physical prompt. Wrote it up in this week's What's 🔥 #512 full post in comments below and other top news from the week! And thanks to Uare.ai for automatically creating this summary video!
-
Ed Sim shared this🤯 Our ChatGPT moment for robots 🤖 is here. Generalist just launched GEN-1.5, a one-shot learner. Robot see, robot do - call it a "physical prompt" It just watches you perform a task and just figures it out and generalizes and just does it - watch below. Sounds simple but extremely difficult. This is a huge deal because robots can now learn brand-new physical tasks from just one short human demo in seconds and immediately generalize to variations in the real world. Traditional robots need extensive retraining or programming for each skill; GEN-1.5 turns that into a near-instant “copy-paste” capability. It marks a major step toward general-purpose robot intelligence that works across many unpredictable everyday situations. Still a lot more improvements to be made but now we all know what is possible! Congrats Pete Florence, Andy Zeng, Andrew Barry and Evan Morikawa and the entire Generalist team. boldstart ventures is a proud investor from inception Ellen Chisa!
-
Ed Sim shared this🤯 This should not be possible yet. You do the task with your own hands. The robot watches. Then it does it with its hands. Different body. Same skill. Seconds later. That’s GEN-1.5 crossing the embodiment gap. boldstart ventures proud investor from inceptionEd Sim shared thisIntroducing GEN-1.5, a one-shot learner. It can learn new tasks in a few seconds. Show it what to do, and it generalizes. GEN-1.5, our latest embodied foundation model, can learn new tasks prompted with 3 - 12 seconds of a single demonstration, no gradient updates or fine-tuning. We refer to this as "physical prompting." It generalizes prompted behaviors to new situations, recovers from mistakes, and improvises new strategies to reach the same goal. For few-shot learning, it can adapt to new physical tasks in as few as 1 - 10 gradient steps on 1 - 5 minutes of data (~10 - 50 demonstrations). Experiments across 10 diverse tasks show 59% average success with one-shot physical prompting. With few-shot learning, performance rises to 83% via 10 gradient steps on 5 minutes of data per task. These capabilities appear to emerge directly from pretraining on large amounts of physical interaction data. We did not explicitly train for this; GEN-1.5 does it out of the box. Although the tasks are simple and success rates are modest, it’s the first model we know of that exhibits the general ability to learn a wide range of dexterous closed-loop physical tasks from just one or few demonstrations. This accelerates reaching a base level of competence for new skills that can be subsequently refined towards mastery. GEN-1.5 has been training continuously for over 8 months. We left it running because every metric we tracked kept improving with the engine: absorbing more data, scaling more efficiently, boosting post-training, and compounding step-change improvements through algorithmic advances. To us, GEN-1.5 represents a new frontier of generality — one that challenges our own understanding of how these models behave when pretrained at a scale of physical interaction data few thought possible without shortcuts. We do not yet see where this asymptotes.
-
Ed Sim reposted thisA $1.1B AUM VC just put his brain on call for founders. 24/7. Sounds impossible. Ed Sim just made it real. His latest note says the venture market has gone barbell. Inception or scale. The middle is death. His advice to founders: control your own destiny. So he did. Ed built his own Individual AI, trained on his real expertise, and now any founder can pressure-test their raise against his actual read, on demand. Fund-grade judgment, on tap, and still entirely his. He told founders to own their destiny. He just did it with his own. Go put your deck in front of his AI 👇
-
Ed Sim shared thisbeen enjoying building my own AI with Uare.ai - i control the data, who can access and for those interested, you can even monetize your expertise just pointed my latest newsletter and here's the short version - will only get better if you want feedback on your deck on the current market go to my AI and chat with me here: https://lnkd.in/gPTiePx6 newsletter from this past week in comments:
-
Ed Sim reposted thisEd Sim reposted thisThe half-life of a moat in cybersecurity is getting shorter. That was the thread running through four conversations I had with investors over the past month, and none of them set out to make the same point. Sid Trivedi of Foundation Capital walked me through the firm's $4.6 trillion Services-as-Software thesis and what it does to security work. His read is that the jobs change rather than disappear. He also cautioned that big raises create growth expectations capital alone cannot meet, with Palo Alto's $25B CyberArk deal and Alphabet's $32B Wiz deal as the backdrop. Ed Sim of boldstart ventures has been investing at inception for 30 years, with roughly a third of Boldstart's portfolio in cyber. His line was that the world needs more cybersecurity, but we don't need all the cybersecurity companies we have right now. He also makes the case that agentic identity is a real category rather than a feature, built on identities that are ephemeral, dynamic and auditable. Jon Sakoda of Decibel Partners framed it as "AI is really only magical to the extent you have a magic power." AI lowers the barrier to starting a company, so the differentiation has to come from the domain expertise underneath. Decibel led a $100M seed into Ent, and he calls endpoint the Super Bowl of cyber. Chenxi Wang, Ph.D. Wang of Rain Capital focused on attacker economics. Vulnerability discovery is becoming a continuous output of computation, which compresses the window between discovery and exploitation from weeks to hours. Taken together, defensibility no longer comes from features. It comes from proprietary data, distribution, domain depth and execution, and that applies to practitioners choosing tools as much as to founders building them. I pulled the full set of takeaways together here👇 https://lnkd.in/eA5tKKhh
-
Ed Sim shared thisThe middle of venture is disappearing. AI startups are getting funded at Inception or Scale and workflow versus infrastructure companies face very different tests. What investors need to see, and how founders caught in the middle can control their destiny: All in my latest issue of What's 🔥 in AI/Infra #511 full post in comments:
Ed Sim replied to a comment
1d
Speed. Adaptability
Ed Sim replied to a comment
2d
Adam Marchick my take is both will be true - will be pockets of death and growth but i do believe that deterministic layer with encoded workflows and rules are stickier than we think. But let’s see how these quarters roll out in future.
-
Ed Sim liked thisEd Sim liked thisSemiAnalysis just published the results of its security testing across neocloud providers - the new generation of cloud companies built around GPU and AI infrastructure. The picture is concerning. These providers are growing extremely quickly, but many of the security controls around tenant isolation, infrastructure access, reprovisioning, and shared management systems have not matured at the same pace. And importantly, most of the issues SemiAnalysis found are not at the AI model layer. They are much lower in the stack: in the infrastructure itself. Tenants could reach systems they should never have been able to access. Network segmentation was sometimes too weak. Machines were reused without fully resetting every layer. Shared monitoring and management systems could expose data across customer boundaries. This is the reality of neoclouds today, and both providers and their customers need to pay attention. We recently published FORGE, the Top 10 Data Center and AI Infrastructure Security Risks, and the overlap with SemiAnalysis’s findings is striking: • Exposed or weakly protected management interfaces • Network and interconnect misconfiguration • Weak tenant isolation • Infrastructure state persisting between customers • Over-privileged shared management systems • Slow patching and remediation cycles GPU scarcity makes this even harder. Customers often cannot choose a provider purely based on security posture - they take the capacity they can get. We built FORGE with security and infrastructure experts from Dell Technologies, Nebius, Google, IBM, Roblox, and others. Use it as a practical baseline to assess a neocloud’s security posture, identify gaps, and ask the hard questions before renting capacity. SemiAnalysis's piece: https://lnkd.in/deTDV_aq FORGE: forge-framework.io #AIInfrastructure #Neocloud #DataCenterSecurity #GPUInfrastructure
-
Ed Sim liked thisEd Sim liked thisFor the past 30 years, my father has spent almost every Friday afternoon in the same place: Tel Aviv’s Drum Beach. It is a place that perhaps could only exist in Israel. Yet amid the headlines that dominate coverage of my homeland, I suspect its story will surprise you. As my dad approached his 80th birthday, I began thinking about how to capture this extraordinary part of his life. I reached out to my favorite Israeli journalist and storyteller, Yigal Mosko of Channel 12 News. Yigal was skeptical. What story could possibly be hiding there in plain sight, right under everyone’s noses? Surely, this annoying Israeli expat was simply trying to get his father on television for a milestone birthday. What he found was something much bigger. The film shows why my father counts down the days until Friday - and how some of Israel’s most unlikely contrasts can coexist and create something beautiful. Engineers and lawyers play alongside people living on society’s margins; Jews alongside Arabs; migrant workers alongside native-born Israelis; parents, veterans (including some carrying the wounds of PTSD), recovering addicts, tourists, and passersby. Every Friday, the circle gathers - summer or winter, peace or wartime. Sometimes its soundtrack is the crashing waves and a cool Mediterranean breeze; sometimes it is sirens and rockets from Gaza or Iran. They barely speak - over the drums, that would be nearly impossible. And even after decades together, many still know almost nothing about one another. The music can be noisy, messy, and out of sync. Then, almost inexplicably, the individual patterns find one another and become something whole. No one asks where you came from, what you believe, or what you do for a living. The only question is whether you are willing to contribute - a voice, a rhythm, or a dance. As someone who has spent much of his career building teams, the lesson I take away is that harmony does not require everyone to play the same beat. It requires us to listen, respond, and make room for the beats of others. I translated and subtitled this beautiful story so that more people could experience the community that has meant so much to my father and has quietly shaped how I understand belonging. https://lnkd.in/gKUjuiGgYigal Mosko - Tel Aviv's Drum Circle 2026 (Captioned)Yigal Mosko - Tel Aviv's Drum Circle 2026 (Captioned)
-
Ed Sim liked thisEd Sim liked thisUnpopular opinion: LinkedIn should be filled with steak & BBQ pics on Saturdays 🥩🔥🦐
-
Ed Sim liked thisEd Sim liked thisLast month, a swarm of AI agents attacked Hugging Face’s systems. It was a preview of a new paradigm for cyberattacks: autonomous, parallelized, and persistent. Cyber defense now needs to be built for that same reality. As AI makes it faster and cheaper to find, chain, and exploit vulnerabilities, defenders must match that speed with continuous testing, autonomous remediation, and verification that fixes actually work. That is why Snyk has joined more than 100 organizations, including OpenAI , Anthropic and Factory, in calling for a global surge in cyber defense. The letter calls for specific action: - Make cyber defense an immediate leadership priority - Continuously test defenses against frontier cyber capabilities - Fix vulnerabilities in the backlog and verify that the fixes work - Put capable defensive AI in the hands of teams protecting essential services The defining challenge is putting AI-powered defense to work at scale before attackers industrialize AI-powered offense. At Snyk, we continue to ship AI security capabilities to help our customers do exactly that. There has never been a more urgent time to build in cybersecurity.
-
Ed Sim liked thisEd Sim liked thisWhen you need to restrict AI usage to enterprise-provisioned accounts in approved tools like Claude, ChatGPT, CoPilot, or Gemini, dope.security makes it possible in minutes.
Experience & Education
Licenses & Certifications
Other similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentAdd new skills with these courses
-
52m
Enterprise AI Solutions with AWS: Amazon Q Business, Bedrock Knowledge Bases, and SageMaker MLOps
-
13m
A Standalone Project: Build a Program to Encrypt and Decrypt Text Messages Using an Encryption Algorithm to Protect Data from Unauthorized Access
-
1h 10m
Advanced AI Analytics on AWS: Amazon Bedrock, Q, SageMaker Data Wrangler, and QuickSight