Get tomorrow's brief in your inbox
Today: Anthropic's Claude dominates enterprise agent platforms at 40% market share (double the nearest rival), but 71% of deployed "agents" are still chatbot wrappers. Indian coding startup Emergent hits unicorn status with $120M ARR selling "engineering teams in a box" to small businesses. xAI open-sourced 840,000 lines of Grok Build code after backlash over uploading SSH keys and password databases to Google Cloud.
Enterprise AI has an orchestration problem masquerading as a platform war. According to VentureBeat's survey of 101 enterprises, Anthropic's Claude leads agent orchestration platforms at 40% adoption, more than double Microsoft (18%) or OpenAI (13%). Companies choose based on "model gravity" (native alignment with state-of-the-art models) and judge success by multi-step task completion reliability.
But here's the gap: 71% of surveyed enterprises admit that a quarter or fewer of their deployed "agents" are actually multi-step orchestrated workflows. The rest are single-prompt chatbot wrappers dressed up with the "agent" label. Only 10% have crossed the halfway mark toward true orchestration.
For small business owners, this means the enterprise AI playbook is being written right now, and the leaders are investing heavily in tooling (34% of budgets) before they have workflows to run on it. If you're evaluating AI platforms, prioritize reliable task completion over platform features you won't use for 18 months. Ask vendors for proof of multi-step execution, not demos of what's theoretically possible. And if you're being sold an "AI agent," ask whether it requires human approval at every step. If it does, you're buying a chatbot, not an agent.
Reelful turns camera rolls into social videos automatically. Former Snapchat ML engineer Kate Deyneka built an iOS app that takes your photos and video clips and outputs polished TikTok/Instagram Reels with AI voiceover, captions, music, and effects. You enter a prompt describing the story, clone your voice with a 30-second sample, select media, and Reelful handles scripting, editing, and assembly. Still images get animated into short video clips (with AI watermarks). Pricing: $15 for 5 videos, or subscriptions starting at $25/month for 10 videos. Target audience: founders and service businesses that need consistent content but lack editing time or resources. If you run a salon, consulting practice, or local service business and have a camera roll full of customer work you've never posted, this is worth testing. Read more
NetSuite Next brings Ask Oracle AI to North America. Oracle's ERP platform for small businesses is rolling out a natural-language assistant that replaces menu-driven searches and manual report building. You can type "show me top customers with overdue invoices" instead of navigating menus and filters. Ask Oracle can answer questions, generate reports, locate records, and complete routine tasks using plain English. The update also introduces AI tools for bank reconciliation, financial close, pricing, and recurring revenue workflows. NetSuite is positioning this as an extension of existing systems, not a reimplementation. Admins can test in a preview environment before enabling it in production. The practical value depends on whether your data is accurate and consistently categorized. If your product names are inconsistent or customer records incomplete, the AI will surface that mess faster than it solves problems. Read more
Zapier leads data integration tools with 9,000+ app connectors. If you're drowning in data spread across unconnected systems, data integration tools can unify siloed information into a single view. Zapier tops the list for no-code automation and near-instant deployment, connecting 9,000+ apps without requiring technical expertise. Other top picks: Informatica for data quality and governance, Fivetran for fully managed connectors, Airbyte for open source extensibility. All have free plans and scalable pricing. The best tools handle connectivity, performance under load, data transformation and quality, and security/governance. If you're manually copying data between spreadsheets and apps, this is the category to investigate first. Read more
Emergent becomes a unicorn selling AI coding to small businesses. Indian AI coding startup Emergent raised $130M at a $1.5B valuation (5x jump in six months) by targeting entrepreneurs and SMBs that run on email, spreadsheets, and messaging apps. Their pitch: "an engineering team in a box" that handles deployment, hosting, testing, and debugging alongside code generation. They've hit $120M ARR with 200,000+ paying customers. Use cases include trucking companies tracking shipments, factories building ERP systems, and property managers creating internal CRMs. One-third of revenue comes from North America, one-third from Europe, and the rest from other markets.
Emergent differentiates from developer-focused tools like Claude Code and Cursor by handling the full stack for non-technical users. The founder acknowledged design remains a weakness (many AI-built websites look similar). If you've been thinking about custom software but lack budget for developers, tools like Emergent, Replit, and similar platforms are now viable alternatives. Expect to trade aesthetic polish for speed and cost savings. Read more
Answer Engine Optimization (AEO) is the new SEO. McKinsey reports 50% of consumers now use AI-powered search, and 70%+ rely on it to gather information. This means your brand needs visibility inside AI-generated answers on ChatGPT, Perplexity, and Google AI Overviews, not just page-one Google rankings. Two platforms are competing for the AEO market: Peec AI (simple, fast insights, competitive benchmarking) and Profound AI (deeper analytics, enterprise integrations, optimization workflows). Both track brand mentions, citations, sentiment, and competitive visibility across AI answer engines. If you're investing in content marketing, you need to know whether AI tools are citing your brand and in what context. Start by auditing your brand presence in ChatGPT and Perplexity for key search terms in your industry. Read more
xAI open-sourced Grok Build after uploading users' home directories. Grok's CLI tool faced severe backlash when users discovered it uploaded entire directories (including SSH keys, password manager databases, documents, photos, videos) to xAI's Google Cloud buckets without clear consent. One user reported running it in their home directory and watching it upload everything. xAI disabled the feature, deleted all retained data, and released the entire codebase (844,530 lines of Rust) under Apache 2.0 license. The code includes tool implementations ported from Codex and Claude (bash, edit, glob, grep, read, write, todowrite) and a Mermaid diagram terminal renderer. Data retention is now off by default. If you tested Grok Build before July 12th, your data was uploaded and has now been deleted. This is a cautionary tale about early-beta AI tools and what "data collection" can mean in practice. Read more
Microsoft trains sales team to talk down OpenAI and Anthropic. At an internal meeting, Microsoft executives outlined a plan for salespeople to negatively compare AI products from OpenAI, Google, and Anthropic to its own in-house models. The pitch: Microsoft sells "the full end-to-end system" while competitors sell "parts." One executive reportedly noted Anthropic's Claude was "slower and less accurate" than Copilot within Microsoft Office apps and lacked proper security integrations. This is notable because Microsoft has long depended on these companies' models to power its own products. Recent reports indicate Microsoft has been swapping out OpenAI and Anthropic models in Word and Excel in favor of its own models as a cost-cutting move. For businesses evaluating Microsoft 365 Copilot, ask which models power which features and whether that's changing. Read more
OpenAI releases $230 keyboard for Codex agents. OpenAI is entering the hardware market with the Codex Micro, a $230 light-up keyboard designed to manage AI coding agents. It includes "Agent Keys" showing agent status, customizable Command Keys for frequent actions, a joystick for workflows, and a dial adjusting agent "reasoning level" (computing power per task). This is a limited-run collaboration with Work Louder, likely more novelty than mass-market product. More significant: Bloomberg reports OpenAI is developing a portable, screenless smart speaker with "mechanical elements that can move on their own," designed by former Apple engineers. Apple is currently suing OpenAI for trade theft, alleging the company extracted confidential information to develop hardware. For now, the $230 keyboard is available for purchase if you want a physical command center for agent work. Read more
Three US states align on AI safety legislation. California, New York, and Illinois have passed frontier AI safety legislation creating a common framework: documented safety frameworks with risk assessments, reporting of serious safety incidents, and governance through independent audits. OpenAI calls this "reverse federalism" where states establish a shared direction that converges toward a national standard. This could mean a de facto national framework emerges from state-level coordination before federal legislation passes. For businesses building on frontier models, expect compliance requirements around safety documentation and incident reporting to standardize across these three states first. Read more
RAG beats fine-tuning for most production use cases. If you're building an AI app and wondering whether to use retrieval-augmented generation (RAG) or fine-tuning, choose RAG unless you have a specific reason not to. RAG gives your model access to external knowledge at runtime without retraining. Fine-tuning retrains the model to behave differently. RAG wins when you need current information, proprietary documents, or frequently changing knowledge. Fine-tuning wins when you need consistent output format or specialized task performance. Cost tradeoff: RAG shifts costs to execution (embeddings, vector database, retrieval per request), while fine-tuning front-loads costs into data preparation and training. Frontier cloud models generally don't support fine-tuning. For most small business use cases (customer support bots, internal knowledge search, document Q&A), RAG is faster to deploy and easier to maintain. Read more
Check your AI visibility with AEO tools. Before investing more in content marketing, audit how your brand appears in AI-generated answers. Open ChatGPT, Claude, and Perplexity. Search for key terms in your industry and see whether your brand gets cited. If you're invisible, your content isn't reaching the 50% of consumers now using AI search. Tools like Peec AI and Profound AI automate this tracking, but you can start manually today. Search for "[your industry] best practices," "[your service] recommendations," or "[your product category] comparison" and note which brands AI tools cite and in what context.
This week's pattern: enterprises are consolidating onto major AI platforms faster than they're building workflows to run on them. The Claude-led orchestration race is real, but the "agents" most companies have deployed are single-prompt chatbots with new branding. Small businesses have an advantage here because they can skip the enterprise playbook entirely and adopt proven tools (Emergent for custom software, Reelful for content, NetSuite's Ask Oracle for ERP queries) that solve specific problems today.
Watch for more state-level AI safety legislation creating compliance requirements around documentation and incident reporting. And expect the AEO category to mature quickly as brands realize half their potential customers are searching via AI tools that may or may not cite them.