Get tomorrow's brief in your inbox
Today: Qodo raised $70M to verify AI-generated code as trust issues mount. Zapier's model routing battle with Gumloop shows the agent automation market splitting. Americans are adopting AI faster but trusting it less, with 76% saying they rarely or only sometimes trust AI outputs.
The verification gap is the new AI bottleneck. As AI coding tools generate billions of lines of code each month, Qodo just raised $70 million betting that verification, not generation, will define software's next phase. The New York startup builds AI agents for code review, testing, and governance, and recently ranked #1 on Martian's Code Review Bench, scoring 64.3% (10 points ahead of the next competitor and 25 points ahead of Claude).
The problem is real. A recent survey shows 95% of developers don't fully trust AI-generated code, yet only 48% consistently review it before committing. Faster output doesn't mean reliable or secure software. Qodo focuses on how code changes affect entire systems, factoring in organizational standards, historical context, and risk tolerance. Co-founder Itamar Friedman explains: "Code generation companies are built around LLMs. But for quality and governance, LLMs alone aren't enough. Quality is subjective. It depends on organizational standards, past decisions, and tribal knowledge."
If you're using AI coding assistants (GitHub Copilot, Claude Code, Cursor), this is the pattern worth watching. The tools making code faster are useful, but the ones making it trustworthy are becoming essential. Source
Zapier Copilot vs. Gumloop: The agent automation split. Zapier published a direct comparison positioning itself against Gumloop's agent-first approach. Gumloop rebuilt around AI agents as the primary interface, with workflows playing a supporting role. Zapier treats agents as one layer of a broader orchestration platform, with 8,000+ integrations, chatbots, tables, forms, and canvas for process mapping. Gumloop offers agents, workflows, and basic forms with 100+ integrations.
The real difference is philosophical. Gumloop wants you to build agentic workflows. Zapier wants agents to work across the apps your teams already use. For small businesses, the question is simple: do you need specialized agent tools, or do you need agents that connect to your CRM, accounting software, and project management apps? If you're already in the Zapier ecosystem, their approach likely makes more sense. If you're building agent-first processes from scratch, Gumloop may be worth exploring. Both platforms offer AI assistants that can create workflows from natural language prompts. Source
Zapier MCP expands Codex to 8,000+ apps. OpenAI's Codex, the agentic coding environment, now connects to Zapier's directory via MCP (Model Context Protocol), giving developers governed access to thousands of actions across business apps. Example workflows include filing GitHub issues from failed test runs, turning Jira tickets into scoped implementation plans, and posting PR summaries to Slack before code review. If you're running engineering teams, this is a way to automate the paperwork around development without leaving your coding environment. Source
LiteLLM ditches Delve after credential-stealing malware incident. LiteLLM, the popular AI gateway used by millions of developers, announced it's dropping compliance startup Delve and redoing security certifications with Vanta after its open source version fell victim to credential-stealing malware. Delve has been accused of generating fake compliance data and using auditors who rubber-stamped reports. The lesson for small businesses: security certifications are only as good as the auditor behind them. If you're evaluating AI tools, ask who did the audit and whether it was independent. Source
Mr. Chatterbox: A Victorian LLM trained on public domain text. Trip Venturella released Mr. Chatterbox, a language model trained entirely on out-of-copyright British Library texts from 1837 to 1899. The 340 million parameter model is tiny (2.05GB) and runs locally using the LLM framework. Honestly, it's weak. Responses feel more like a Markov chain than a useful assistant. But the concept is fascinating: a model with zero copyright issues, trained exclusively on public domain data. Simon Willison suggests we'd need 4x the training data to get something conversationally useful. For now, it's a proof of concept, not a productivity tool. Source
Small businesses waste hours scheduling meetings manually. A Zoho Bookings survey of 1,700+ respondents (254 from US small businesses) found 72.8% still use email to schedule meetings, 57.1% use phone calls, and 51.6% check availability via email and messaging. Only 34.7% use dedicated scheduling software. The result: 46.5% schedule meetings in under an hour, but 28.4% take 1-6 hours, 13.8% take 6-12 hours, and 8.3% take up to 24 hours. The biggest time drain is back-and-forth scheduling (46.1%), followed by no-shows (34.7%) and double bookings (31.9%).
Here's the opportunity: Small businesses using scheduling software report benefits in automation, meeting management, employee productivity, team coordination, and customer experience. Yet only 18.2% say their tool meets their needs. The top requests are more AI capabilities (40.9%), better integrations (39.8%), and improved customer support (37.5%). If you're still emailing to find meeting times, switching to Calendly, Cal.com, or Zoho Bookings will save hours weekly. Source
Quo (formerly OpenPhone) scales with Zapier integration. Quo, a VoIP business phone platform, has seen 17% quarter-over-quarter growth in active Zapier integration users since 2022. Customers using Zapier generate 70% higher annual revenue than those who don't. The top use case: connecting Quo to CRMs. The pattern here is worth copying. If you sell software or services to small businesses, investing in automation integrations drives retention and revenue. Quo embedded Zapier's Partner API into their product, created onboarding emails highlighting the integration, and partnered with Zapier Experts to help customers set up custom workflows. Source
ScaleOps raises $130M to cut cloud costs by 80%. The startup builds software that automatically manages and reallocates computing resources in real time, targeting companies struggling with idle GPUs, over-provisioned workloads, and climbing cloud costs. ScaleOps claims its platform reduces cloud and AI infrastructure costs by as much as 80%. The company serves enterprise customers running Kubernetes-based infrastructure and just hit an $800 million valuation. If you're spending $10k+ monthly on cloud compute, this is the category to watch. Competitors include Cast AI, Kubecost, and Spot. Source
Americans adopt AI faster but trust it less. A Quinnipiac University poll of 1,400 Americans found 76% trust AI rarely or only sometimes, while just 21% trust it most or almost all of the time. Yet adoption is rising: only 27% have never used AI tools (down from 33% in April 2025). The contradiction is striking. Half use AI for research, many use it for writing and data analysis, but only 21% trust the results. Meanwhile, 55% say AI will do more harm than good in their daily lives, and 70% believe AI will cut job opportunities. Gen Z is the most pessimistic, with 81% foreseeing job losses.
What this means for small businesses: your customers are using AI but don't trust it. If you're building AI features into products or services, transparency and verification layers matter more than speed. People want to know how the AI reached its conclusion and whether a human reviewed it. Source
Okta CEO "paranoid" about the SaaSpocalypse. Todd McKinnon, CEO of the $14 billion identity management company, recently told investors he's "paranoid" about AI agents replacing SaaS tools. The concern: why pay SaaS fees when you can vibe-code your own tools? Okta's response is betting on "agent identity" as a new market. The idea is that AI agents inside corporations need access and security credential management just like people do, especially with tools like OpenClaw handing credentials to agents. McKinnon suggests a "kill switch" at the agent level as one solution. The deeper question: if AI can build software, what does it mean to run a software company? Source
Mistral raises $830M in debt for Paris data center. French AI lab Mistral secured $830 million in debt financing to build a data center near Paris powered by Nvidia chips, operational in Q2 2026. The company previously announced $1.4 billion for Swedish AI infrastructure and aims for 200 megawatts of compute capacity across Europe by 2027. CEO Arthur Mensch: "Scaling our infrastructure in Europe is critical to empower our customers and ensure AI innovation and autonomy remain at the heart of Europe." For context, this is the European response to US hyperscaler dominance. Source
Automate lead follow-up with VoIP + CRM workflows. If you're using a virtual phone system (8x8, Dialpad, Quo), connect it to your lead generation tool via Zapier. When new leads come in (Facebook Lead Ads, Typeform, contact forms), automatically send an SMS or create a CRM contact. This eliminates the "fell through the cracks" anxiety and ensures every lead gets touched within minutes. Pre-built templates exist for most popular platforms.
File GitHub issues from failed test runs using Codex + MCP. If you're running CI/CD pipelines, connect Codex to GitHub via Zapier MCP. Prompt: "Read the most recent test output. For each failing test, check if an open issue exists. If not, create one titled '[Bug] [test name]' with the error message and file path, labeled 'bug' and assigned to me." This turns test failures into tracked work without manual ticket creation.
The verification layer is becoming the AI product. Whether it's Qodo validating code, scheduling software cutting wasted hours, or trust metrics showing up in user surveys, the pattern is clear: speed alone isn't the value proposition anymore. Businesses want AI that works reliably, integrates with existing tools, and can be verified. Watch for more "AI verification" startups and features that add transparency layers to AI outputs. If you're building or buying AI tools, ask: how do I know this is correct? The companies answering that question win.