Zoho CEO: “AI Agents Aren’t SaaS. Stop Treating Them That Way.”

🤖 AI Agents

Zoho CEO: “AI Agents Aren’t SaaS. Stop Treating Them That Way.”

What happened: Zoho CEO Mani Vembu published a thread on X that cut through the AI agent hype with surgical precision. Responding to The Browser Company CEO Josh Miller’s observation that “most people are still using ChatGPT and Claude as a glorified Google + Grammarly,” Vembu argued the bottleneck isn’t model capability — it’s onboarding friction. “The problem isn’t the models. It’s time to value,” he wrote. “SaaS delivered value almost immediately through self-service. AI agents require ‘agent sitting’ before they become useful. You have to teach them your business, your data, and your workflows.”

Why it matters: This is the clearest articulation yet of why AI agents haven’t crossed the chasm despite genuinely impressive demos. Vembu’s core insight — “even for the same use case, the same agent rarely works out of the box for two different businesses” — explains why enterprise adoption is stalling at the pilot stage. Every business has bespoke processes, tribal knowledge, and undocumented workflows. An agent that can write code for one company flounders at another until it’s painstakingly trained. The SaaS playbook of “sign up and go” simply doesn’t apply.

What’s next: Zoho’s answer — unveiled at ZohoDay 2026 — is AppOS, a unified platform linking 50+ apps and agents with privacy and governance built in. The bet is that agents need deep integration into the existing software stack, not standalone deployment. If Vembu is right, the winners in the agent era won’t be companies with the smartest models — they’ll be the ones that solve the onboarding problem. Watch for “agent sitting” to become a real job category in 2026-2027.

Source: India Today

⚡ Building on LLMs Is “Closer to Biology Than Software Engineering.” Anthropic’s Boris Cherny, creator of Claude Code, says every new model generation “behaves differently — it has a slightly different personality, and you have to take the time to get to know it.” His rule of thumb: system prompts and code harnesses are disposable by design; evals last 1-3 model generations, then saturate and get thrown away. The implication? Agent infrastructure built around a specific model’s quirks has a short half-life. OfficeChai

⚡ Only 18% of Enterprises Are AI “Pacesetters.” A massive ServiceNow–Oxford Economics survey of 4,470 executives found the vast majority of organizations are still stuck in experimental AI mode. The 18% generating real business value do one thing differently: they embed AI into core business processes via integrated platforms (like the new ServiceNow–Microsoft Copilot tie-in for 400M+ M365 users) rather than bolting it onto the edges. Technology Record

📰 AI News

🧬 AI Designs 16 Brand-New Viruses From Scratch — Stanford’s Landmark Science Paper. Stanford researchers used an AI model called Evo to design entirely novel, functional viruses (bacteriophages) that don’t exist in nature. All 16 replicated successfully in lab tests targeting E. coli. The team explicitly excluded human pathogen data, but biosecurity experts warn this raises “urgent biosafety questions.” On the upside: AI-designed phages could become a powerful weapon against antibiotic-resistant superbugs. The Star / BBC

💊 Anthropic’s Drug Discovery AI Pulled After FDA Dispute. Anthropic unveiled an AI model on August 4 that can design drug candidates 10x faster than existing methods — potentially shrinking the discovery phase from years to months. But the model was temporarily pulled from public access amid a regulatory standoff with the federal government. The bottleneck: even if AI finds the molecule in weeks, clinical trials still take years. The FDA now faces pressure to modernize without repeating a thalidomide-level mistake. The Tech Edvocate

💰 European AI Startups Capture 21.2% of July’s €8.6B Funding Haul. AI was the dominant sector in European tech funding for July 2026, pulling in €7.7 billion across the month. Germany led with €3.5B over 48 deals, headlined by defense AI startup Helsing’s $1.8B Series E at an $18B valuation. Exit activity also accelerated — 51 transactions, up from 39 in June — signaling a maturing AI ecosystem beyond just funding rounds. Tech.eu