Prasenjit Sarkar

Welcome Salesforce Headless 360: No Browser Required! Our API is the UI. Entire Salesforce & Agentforce & Slack platforms are now exposed as APIs, MCP, & CLI. All AI agents can access data, workflows, and tasks directly in Slack, Voice, or anywhere else with Salesforce Headless 360. Faster builds, agentic everything. 🚀 #Salesforce #Agentforce #AI venturebeat.com/ai/salesforce-…

Introducing Claude Design by Anthropic Labs: make prototypes, slides, and one-pagers by talking to Claude. Powered by Claude Opus 4.7, our most capable vision model. Available in research preview on the Pro, Max, Team, and Enterprise plans, rolling out throughout the day.

Introducing web-agent, an open framework for building web agents 🔥 Build AI agents that search, scrape, and interact with the web - powered by the same architecture behind our /agent endpoint. 100% open source. Bring any model. Anthropic, OpenAI, or your own.

We've redesigned Claude Code on desktop. You can now run multiple Claude sessions side by side from one window, with a new sidebar to manage them all.

Introducing Fire-PDF, our new Rust-based parsing engine 🔥 - Convert PDFs into markdown 5x faster - Extract full tables and preserve formulas - Zero config required

What an incredible video of the Artemis II crew capsule coming down to earth at incredible speed

Introducing Claude Managed Agents: everything you need to build and deploy agents at scale. It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days. Now in public beta on the Claude Platform.

You can now run Cursor on any machine and control it from anywhere. Kick off agents from your phone to run on your devbox.

1/ today we're releasing muse spark, the first model from MSL. nine months ago we rebuilt our ai stack from scratch. new infrastructure, new architecture, new data pipelines. muse spark is the result of that work, and now it powers meta ai. 🧵

Maybe the Delve founders are genuine psychopaths. Maybe they set out to defraud people from day one. But they're 21. Both of them. Basically kids. More likely: they got into YC, got surrounded by people telling them to move fast and break things, grow at all costs, fake it till you make it, and so on. They wanted to impress their batchmates, their partners, the alumni network, their parents. So they pushed too hard and broke things they shouldn't have broken. YC selects for this. They want young naive founders who are aggressive, ambitious, a little reckless. They celebrate the ones who bend the rules and win. When a 21 year old bends the rules and loses, suddenly it's a "trust" issue and they get a cold three-sentence farewell on Bookface. No mentorship. No "hey, you're heading in a dangerous direction." No community stepping in before it got this far. Just "we asked them to leave, we wish them well." That's not a community. That's a machine that takes credit for your wins and disowns you when you mess up.

@kocalars Getting booted out has got a new name “parted ways”

YC and Delve have parted ways. I still remember the day we took our YC interview at MIT. We’re so grateful to the community and every founder friend we’ve made. We'll continue to support every young founder striving to make the world a better place.

Computer use is now in Claude Code. Claude can open your apps, click through your UI, and test what it built, right from the CLI. Now in research preview on Pro and Max plans.

Introducing Critique, a new multi-model deep research system in M365 Copilot. You can use multiple models together to generate optimal responses and reports.

$285 billion disappeared from SaaS valuations in 48 hours. Everyone blamed Anthropic for “SaaSpocalypse”. They were wrong. On January 30, Anthropic dropped 11 open-source plugins for Claude Cowork. Four days later, $285 billion in market cap was gone. LegalZoom down 20%. Thomson Reuters down 16%. The JPMorgan Software Index down 7% in a single session. Wall Street called it the SaaSpocalypse. I think that's the wrong diagnosis entirely. Here's what's actually happening. SaaS is not being killed by a model release. It's being killed by its own failure to evolve. The difference matters enormously, because one is a threat from outside and the other is a self-inflicted wound. Take the "vibe coding kills SaaS" argument. Yes, you can describe requirements in plain English today and have a working app in minutes. I've seen founders replace $300/month tools overnight. But a weekend prototype is not a system that survives 10,000 concurrent users, a SOC2 audit, 17 legacy integrations, and a 3am outage. The people celebrating are coding yesterday's legacy systems. The right question is what software should feel like when AI is the operating system from day one. Then there's the agent argument. The idea that Claude sitting on top of Salesforce or HubSpot kills those platforms misses something fundamental about architecture. Aaron Levie put it cleanly: "We'll have about 100 times more, maybe 1,000 times more, agents than we have people. So you'll have way more users of that software system as agents." The agent is the brain. The SaaS platform is the memory. One does not function without the other. So who actually kills SaaS? Not a specific model. Not a plugin drop. Bain & Company found that nearly 35% of SaaS vendors have already increased per-seat pricing with bundled AI features, while another 35% are moving to hybrid usage-based models. The per-seat era is ending regardless. What replaces it is the question. The real threat is whoever reimagines the solution from ground zero. Not an AI wrapper on legacy architecture. Either an incumbent brave enough to disrupt itself, or an AI-native startup that treats intelligence as the foundation, not the add-on. Gartner said it plainly after the selloff: "Predictions of the death of SaaS are premature. What this exposes is how much day-to-day knowledge work remains manual." The panic was misplaced. The disruption is real. Those are two different things. We're moving from software you use to software that does. That transition won't be overnight. But the blueprint is changing fast — and the companies that will die aren't the ones being disrupted by Anthropic. They're the ones still waiting to disrupt themselves.

You can now enable Claude to use your computer to complete tasks. It opens your apps, navigates your browser, fills in spreadsheets—anything you'd do sitting at your desk. Research preview in Claude Cowork and Claude Code, macOS only.

Zuckerberg banned OpenClaw for every Meta employee. Then quietly started building one for himself. Six weeks ago, OpenClaw was a side project by an Austrian developer who came out of retirement feeling bored. Today, Jensen Huang is telling 30,000 people at GTC that every company on earth needs an OpenClaw strategy. That's not a product cycle. That's a regime change. Here's what's actually happening — and what it means for the people closest to the C-suite. OpenClaw crossed 250,000 GitHub stars in 60 days. React took a decade. Linux took longer. It is now, by any measure, the fastest-growing open-source project in the history of the platform. NVIDIA didn't wait. They shipped NemoClaw — an enterprise security layer on top of it — and brought 17 major software companies along for launch week. And Zuckerberg? While Meta's internal policy was banning employees from running OpenClaw on work devices, he was personally testing it. Texting the creator. Losing the talent war to Sam Altman. Then spending $2B on Manus and acquiring Moltbook to build his own version of the same thing. Now the WSJ is reporting he's building a personal CEO agent. The reason is simple. A CEO at Meta's scale makes decisions on maybe 1% of the information that would actually change the outcome. The other 99% dies somewhere between the source system and the pre-read document. It gets filtered by VPs, summarised by chiefs of staff, averaged out by dashboards, and translated into whatever fit on a slide. An agent doesn't filter. It reads everything. Simultaneously. And surfaces what matters before the 8am. Before a 1:1, it already pulled that person's delivery record, their open headcount, their last three missed commitments. When someone asks what happens to a product timeline if 200 engineers get reassigned, it doesn't kick off a two-week strategy exercise. It runs the scenario in minutes. And it never loses institutional memory. The person who remembered why that initiative was killed in 2019 left two years ago. The agent was there. Now scale that beyond the CEO seat. Most large companies run 8–12 people around a single executive just to manage information flow. Multiply that across every SVP with a CoS, every VP with a BizOps partner, every director with an EA. The core loop those roles run: gather, filter, synthesise, route, track. Every step is text in, text out. An agent does that loop in minutes, not weeks. It doesn't need to Slack four teams for the latest numbers. It's already connected to the source. The honest projection: 90% of information-routing roles dissolve over the next five years. What survives is the 10% that can't be reduced to text — reading a difficult room, managing a relationship that's about to break, knowing that "let's revisit" actually means no. Everything else becomes infrastructure. Zuckerberg building this for himself isn't a curiosity. It's a preview of what every board will be asking their CEO about by 2027. What are you building yours on?

Let me explain exactly why the entire $32 billion compliance industry is quietly panicking right now. A YC-backed startup raised $32 million to automate SOC 2 compliance. They just got caught faking 494 audit reports. The mechanism behind the fraud is wild. Delve (@getdelve) launched in YC's W24 batch with a simple pitch: AI that makes compliance effortless. They promised to get startups SOC 2 compliant in a fraction of the time. Insight Partners led their Series A at a $300 million valuation. 1,500+ companies signed up. The growth curve looked like a rocket ship. The reality looked like a sweatshop. In late 2025, an employee accidentally shared an internal Google Spreadsheet with "anyone with the link" permissions. It was indexed by search engines before it could be taken down. That spreadsheet contained Delve's entire audit report generation pipeline. The data is staggering. Out of 494 leaked SOC 2 reports, 493 had near-identical boilerplate. 259 had identical Type II conclusions. They weren't automating compliance. They were automating the illusion of it. The platform generated auditor conclusions before any auditor reviewed the evidence. They partnered with Accorp, an Indian audit firm operating through US shell addresses, to rubber-stamp whatever the system spit out. But the most damning part is how they handled the evidence. When a client's employees didn't complete their security training or background checks, Delve didn't flag it. They faked it. The system generated fake evidence for device security. It sent AI-generated emails impersonating HR departments. They built a feature to forge reality. The companies holding these reports—startups like Incorta, Flow, and Bland.AI—thought they were buying enterprise readiness. Instead, they bought criminal liability. Under HIPAA, holding fraudulent compliance reports can lead to criminal charges. Under GDPR, it's a fine of up to 4% of global revenue. The industry response tells you everything you need to know about the structural vulnerability of self-policing attestation models. SOC 2 has no enforcement mechanism. It is a purely self-regulatory model created by the AICPA. If a report is fraudulent, there is no regulator to compel corrective action or impose fines. The affected parties have to pursue private remedies. The system is designed to optimize for speed, not security. When you incentivize startups to view compliance as a blocker to revenue, and you incentivize auditors to view startups as recurring revenue, you create a market for rubber stamps. Delve just figured out how to mass-produce the stamp. The EU AI Act was designed specifically to prevent this. It requires continuous quality management, not point-in-time screenshots. It mandates market surveillance authorities with real enforcement powers. It structurally separates the platform vendor from the certification body. But in the US, the market is still pricing itself on the illusion of security. The question isn't how one startup managed to fake 494 compliance reports. The question is how many other platforms are doing the exact same thing, just waiting for someone to share the wrong Google Drive link.

Delve raised $32M from YC to be the fastest compliance platform on the planet. 494 leaked audit reports later, we know how they did it. Here's the breakdown: A leaked Google spreadsheet exposed every draft SOC 2 report Delve had produced. 493 out of 494 contained identical boilerplate text. Same sentences. Same grammatical errors. Same nonsensical phrasing. Only the company name, logo, org chart, and signature were swapped in. This is not a minor corner-cutting story. Auditor conclusions and test procedures were pre-written in draft reports before clients even submitted their company description. That is a structural violation of AICPA independence rules, which require auditors to independently design tests and form conclusions. Delve was doing the opposite. All 259 Type II reports shared the exact same outcome: Zero security incidents. Zero personnel changes. Zero customer terminations. Zero cyber incidents. Identical "unable to test" conclusions — across every single client. The probability of that happening in a legitimate audit is basically zero. Their advertised "US-based CPA firms"? Actually Accorp and Gradient — Indian certification mills operating through US shell entities. Compliance theater. A SOC 2 template pack with a SaaS wrapper and a $32M Series A. The client list is where this gets real. Lovable. Bland. Cluely. NASDAQ-traded Duos Edge. Every enterprise deal closed using these reports is now legally exposed. HIPAA violations carry criminal liability. GDPR fines run up to 4% of global revenue. Lovable left the moment the story dropped. Here's the part nobody's talking about. When a client threatened to churn, Delve sent a vCISO to do the work manually, off-platform. That single move tells you everything. If their AI platform actually delivered real compliance, why would manual work be the retention play the moment real stakes appeared? CEO Karun Kaushik's response to the leak called it "falsified claims from an AI-generated email." The response email was itself drafted by AI. The leaked data included private signatures and confidential architecture diagrams. That's not fabricated. That's a data breach on top of a fraud. The business model makes perfect sense once you understand the incentive structure. Compliance is a checkbox. Startups don't want six months of SOC 2 work. They want the badge to close the enterprise deal. Delve gave them the badge without the compliance. Sales price dropped from $15,000 to $6,000 with ISO 27001 and a pen test thrown in the moment a client mentioned a competitor. The margin math only works if the product costs nearly nothing to produce. It does. It's a template. This is what happens when you optimize for speed over trust infrastructure. The pitch was "11x faster than competitors." Enterprise buyers rewarded it. The market rewarded it. YC backed it. Nobody asked how. Delve built a machine designed to make clients complicit without their knowledge. That's the actual product: plausible deniability at scale. Your SOC 2 report looks legitimate. You believe you're compliant. The auditor's name is on it. Until a spreadsheet leaks. The real signal here isn't Delve. It's that compliance-as-a-checkbox was always going to produce this outcome. When the buyer doesn't understand what they're buying, and the seller is rewarded only for closing, the template was inevitable. Every YC compliance startup is under a microscope now. Speed was never the feature. It was the tell.

We just released Claude Code channels, which allows you to control your Claude Code session through select MCPs, starting with Telegram and Discord. Use this to message Claude Code directly from your phone.