Agentic — July 14, 2026

AI Agents Are Moving Into Real Work. What Happens to SaaS? | Sierra AI Co-founder Clay Bavor

Clay Bavor, co-founder of Sierra AI, explains that AI agents represent a new category of software capable of reasoning, decision-making, and autonomous action — and are already delivering measurable results in customer service, mortgage origination, and sales for enterprise clients. Sierra uses a constellation of frontier and fine-tuned models, with supervisory agents acting as guardrails to catch errors in real time, and can deploy a fully functional agent for a major company in as little as 58 days. Bavor highlights Sierra's "outcomes-based pricing" model — where clients only pay when an agent fully resolves a customer issue — as a direct alternative to traditional SaaS seat-based subscriptions, arguing it better aligns incentives and simplifies ROI calculations. He also addresses the "token maxing" trend, acknowledging that while coding agents in particular deliver genuine productivity gains, companies are now being forced to scrutinize AI spending more carefully and will likely begin budgeting token usage the way they budget other employee resources.

I Tried Making $800 in 4 Hours with AI Agents (To See If It Works)

This video documents a creator's attempt to build an AI-powered booking and client management system for a salon using a no-code tool called Base44, framing it as a way to generate $800 in four hours. The system demonstrated includes an AI receptionist agent that answers inquiries, books appointments via Google Sheets and Calendar, sends Telegram reminders to reduce no-shows, and processes deposits through Stripe. The creator walks through the live build step by step, showing how the agent handles the full client flow from initial inquiry to post-appointment follow-up without manual intervention. The video also promotes a paid Base44 masterclass and suggests the same AI agent framework can be applied to other service businesses like clinics, barbershops, and consultants.

Agents A1 Benchmarked vs Qwen 35B - 16GB Local LLM setup

Lukas fra Luke's Dev Lab benchmarker Agents A1 (en 35B Mixture-of-Experts-model fra Intern Science i Q4KM GGUF-format) mod Qwen 35B på et 16 GB VRAM-system. Hastighedsmæssigt er de to modeller næsten identiske, men i hukommelsestesten klarer Qwen sig markant bedre ved at finde data i den fulde 256K-kontekst med færre fejl og renere output. I agentur-benchmarket scorer begge modeller 95%, idet de begge fejler på den samme valutakonverteringsopgave, dog på lidt forskellig vis. Videoen konkluderer, at Agents A1 er en stærk model, men at Qwen 35B generelt performer bedre eller på niveau på tværs af de testede discipliner.

Every Local AI I Run Now Shares ONE Memory | (LLM Wiki + OKF)

The video describes building a shared memory system for local AI models, inspired by Andrej Karpathy's LLM Wiki concept and Google's OKF spec. Rather than burdening the AI with complex memory instructions (which smaller local models struggle to follow), the creator built a separate "librarian" agent that manages a structured, markdown-based memory store — handling queries, updates, and new entries via MCP tools, while the main AI stays focused on its actual task. Key challenges solved include preventing the model from ignoring memory entirely (fixed by injecting a live summary at session start), stopping it from creating disconnected files instead of enriching existing concepts, and handling contradictions and orphaned memory nodes. The result is a fully local, cloud-free memory layer shared across all AI tools, with planned additions like a recent-memory cache and an idle "dreaming" mode that reorganizes and consolidates memories automatically.

agentic ai cringelords

The video is a sardonic critique of social media influencers who post viral "agentic AI" setup videos — typically showing a voice-activated "Jarvis"-style AI giving them polished business reports — which the host argues are staged, impractical, or outright fake. Eric Morrison mocks the broader hype culture around AI agents, pointing out absurdities like agents giving identical reports across multiple videos, inflated metrics, and the general premise that running more agents automatically equals more success. His core argument is that this content spreads a harmful false narrative — that you must automate everything with AI agents or be "left behind" — when in reality checking a dashboard is faster and cheaper than listening to a spoken AI report.