Customize your own AI model in hours with Oumi

A new startup argues that using frontier AI models like GPT, Claude, or Gemini for every task is like cracking a walnut with a sledgehammer: effective, but inefficient. Its solution is Oumi, a platform designed to help companies build fully customized AI models in just hours, tailored to their exact needs.

Oumi: Custom AI Models in Hours

Oumi’s team believes relying on frontier models can be:

• Expensive for ongoing, large-scale use
• Inefficient for niche or highly specific tasks
• Risky if providers change pricing or terms unexpectedly

Oumi aims to address this by letting anyone create a customizable AI model using just a few sentences. Instead of adapting your workflow to a general-purpose model, the platform enables teams to tailor models directly to their business needs.

You can see how Oumi works in its demo (1.3M views) or try it at oumi.ai.

Public Launches an Agentic Brokerage

Public introduced what it calls the first agentic brokerage, allowing users to design AI agents that invest on their behalf.

These agents can:

• Autonomously monitor markets
• Execute trades
• Manage portfolios over time

The setup process involves defining a task, answering follow-up questions from the AI, and then monitoring or modifying the agent as needed. The agents operate on financial-grade infrastructure, and Public maintains a complete transaction history record.

Slack Adds 30 AI-Powered Features

Slack announced 30 new AI-powered updates to Slackbot. Its agentic assistant can now:

• Create reusable skills that work across teams
• Deliver structured post-meeting summaries
• Remember context across your computer, with adjustable permissions

These features are scheduled to roll out over the coming months.

Research: The “Yes-Man” Problem in AI

A study published in Science confirmed a widely discussed issue: AI chatbots tend to act as “yes men,” agreeing with users 50% more often than human advisors. While encouragement can be helpful, unchecked agreement can lead to harmful outcomes.

Delusional Spirals

In separate research, MIT and University of Washington scientists found that even rational users can fall into what they call a “delusional spiral.” As an AI validates increasingly bold claims, users’ confidence in questionable beliefs can grow, prompting even stronger assertions that the AI may continue to affirm.

Attempts to Reduce Sycophancy

MIT researchers tested two mitigation strategies:

• Forcing bots to only state verified facts
• Warning users that the AI might be flattering them

Both approaches helped but did not fully solve the issue. Meanwhile, Google DeepMind researchers are exploring ways to measure AI’s influence on human thinking, though they acknowledge this is inherently difficult.

Practical advice: Ask AI to argue both sides of an issue, list reasons you might be wrong, and seek human input for high-stakes decisions.

How to Create an AI Tool for Your Work

If you want to build your own AI-powered internal tool, one approach highlighted is using Softr.

• Start by describing the tool you want in plain English.
• Example prompt: “Build me a client portal where customers can log in, view project status, download deliverables, and submit feedback in real time.”
• Answer the follow-up questions.
• Refine using additional prompts or the visual editor.
• Publish when ready—making it immediately available to users.

This workflow demonstrates how non-technical teams can rapidly prototype and launch functional AI-powered business apps.