Deepak Gupta's notebook

Cybersecurity has the hardest content audience in B2B, and the reason is structural rather than attitudinal.

Most founders building on someone else's model believe the EU AI Act is the model provider's problem.

In 2025, URIports analyzed the top one million domains and found that 53.6 percent of BIMI-enabled domains had at least one error preventing their logo from displaying, up from 41.8 percent the year before.

Most teams treat LLM output as deterministic.

There is a moment in every B2B deal where it stops being a deal and becomes somebody else's problem.

On July 21, 2026, OpenAI confirmed that one of its own evaluation agents had broken into Hugging Face.

Four unrelated corners of the internet spent 2026 solving the same problem, and mostly did not notice each other doing it.

I have spent years arguing that passkeys are the right answer to phishing.

On August 4, 2026, the UK's AI Security Institute published a finding that attacks identity verification from a direction most platforms have never had to consider.

Enterprise security has spent two years building defenses for AI agents that attack systems on behalf of a human, or, less often, on their own.

CrowdStrike's 2026 Global Threat Report puts average eCrime breakout time at 29 minutes, down from 48 minutes a year earlier, with AI-augmented attacks up 89 percent year over year.

On August 26, 2026, Meta settled with 51 state and territory attorneys general for a guaranteed 12.1 billion dollars over ten years, rising to 17.1 billion if other major platforms adopt the same safeguards.

The Model Context Protocol shipped its largest specification revision since launch on July 28, 2026.

Most cybersecurity vendors build their entire go-to-market around convincing one person: the CISO.

Now that ChatGPT, Google, and Microsoft all sell placement inside AI experiences, marketing teams face a budget question that did not exist eighteen months ago: how much goes to paid AI placement, and how much to earning organic citations?

Security taxonomies get used interchangeably in conversation, and they should not be.

GrackerAI paid WorkOS $625 a month for five enterprise SSO (Single Sign-On) connections.

Every enterprise now manages 109 machine identities for every human one, up from 82 to 1 a year earlier, according to Palo Alto Networks' 2026 Identity Security Landscape report.

Ninety days is enough to fix crawler access, establish a measurement baseline, restructure your ten highest-value pages, and close the largest coverage gaps in your category.

Engineering teams do not resist AI initiatives because they doubt the technology.

Somewhere in your infrastructure right now, a process is deciding on its own which websites to visit.

Only 13% of enterprises have deployed passkeys at scale.

Ninety-five percent of generative AI pilots inside large companies produce no measurable financial return.

CPU cache is a small amount of very fast memory built into the processor itself, sitting between the cores and your computer's main memory.

Every tool in this category sells "AI visibility" and almost none of them measure the same thing.

On 20 July 2026, W3C proposed advancing Web Authentication Level 3 to Recommendation, the last stage of the standards process.

94% of teams running AI agents in production have implemented some form of observability, according to LangChain's survey of 1,300+ practitioners.

The five enterprise CIAM platforms that actually make enterprise shortlists in 2026 are Auth0 (Okta's Customer Identity Cloud), Microsoft Entra External ID, Ping Identity (which now includes ForgeRock), IBM Verify, and WSO2 Identity Server.

You can tell within the first hour of a hackathon whether it was planned or improvised.

Query fan-out is the step where an AI search system takes a single prompt, decomposes it into multiple related sub-queries, retrieves sources for each one independently, and then assembles a single answer from the results.

Dozens of people cancelled their Claude subscriptions this week over a feature that has no effect on their search rankings.

Most founders treat fundraising as a personality test.

India turns 79 today, and I want to talk about what it gave me before any accelerator, investor, or market ever did.

Most AI search statistics circulating in 2026 have no traceable source.

Google published its official position on AI optimization in 2026, and it could not be clearer.

Sit in enough B2B SaaS budget meetings and you notice the same scene repeat.

You have budget approval.

Anthropic's Claude for Startups program does not publish specific dollar credit tiers anywhere on its own program page.

Prompt engineering is overrated.

Adobe closed its acquisition of Semrush in April 2026, a deal valued at roughly $1.9 billion in cash, and framed it explicitly around brand visibility in the agentic AI era.

The four engines that matter retrieve differently, cite differently, and reward different things.

A B2B contact database decays about 22.5 percent a year, according to HubSpot's database decay research, which cites MarketingSherpa's data.

A leaked credential is not one problem.

Every founder I talk to thinks their data is "mostly clean."

I grew up in Rajasthan, built my first products with a team sitting in India, and then spent years learning how to sell software to people in San Francisco and New York who had never heard my name.

For 90 days I tracked how six major AI search engines cite sources when they answer B2B software queries.

In June 2025, a security firm showed that a single email could turn Microsoft 365 Copilot against the company running it.

GEO and SEO share a foundation and diverge at the top.

Every founder I know hits the same wall the first time an enterprise prospect asks for their SOC 2 report.

It is Hacker Summer Camp week.