Urgent.News

What's breaking now, across thousands of outlets.

AI

Before/after: CRM made agent-ready

Let's walk through the same task twice: "Find every enterprise deal closing this quarter and draft a follow-up for each owner." Before: the agent loads the CRM. It reads the pipeline page like a screenshot, guesses which column is "Close date," and tries to click a filter it can't quite see. Each page of results is another full page-load of tokens. It misclicks, loses the filter, starts over. The…

Let's examine the same task of finding enterprise deals closing this quarter and drafting follow-ups for each owner twice.

First, the agent loads the CRM system. They view the pipeline like a screenshot, try to figure out which column represents the Close date, and attempt to click a filter that's difficult to locate. Each page of results requires loading a full page of data. Mistakes are made, filters are lost, and the process has to be restarted. Drafts of follow-ups are typed character by character into the compose box, as there's no other way to proceed.

Now, consider the same CRM, but this time made agent-ready. The agent issues a structured action: list deals, stage = enterprise, close date within Q4. The system quickly returns clean records: name, value, owner, stage, with no extraneous data. Another action drafts the follow-ups against the appropriate templates, and a step ensures all approvals from owners are obtained before anything leaves the building. The same task is accomplished, but the outcomes are vastly different.

The difference lies not in the agent's intelligence, but in what the software allows them to touch. The previous version resembled browser automation trying to mimic a user. The new version establishes a clear contract between the product and the agent, with built-in safeguards. This represents "agent-ready" in practice - not a rebuilt product, but an honest interface atop the existing system.

Which workflow in your stack would you trust an agent with first - the one that mimics human behavior with flawed results, or the one that provides a clear, structured interface? Which would you keep as a human-only domain?

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in AI

The 0.87 problem: when semantic linking makes inconsistent records look connected

A 0.87 match that almost shipped Last quarter I built a small semantic-similarity layer over our CAPA database. The pitch to myself was simple: an engineer files a CAPA, the system surfaces related…

  • Semantic linking creates inconsistent records appearing connected
  • Inflation problem inflates false confidence in similarity scores
  • Three steps: source consistency, human ratification, traceable decision lineage

The Modular AI Supply Chain: Why Autonomous Agent Skills Need Pre-Install Security Scanning

As autonomous AI agents shift from experimental scripts into production systems, the way software engineering teams extend agent capabilities has fundamentally changed.

  • Autonomous AI agents use modular Skill Bundles to add functionalities.
  • Traditional SAST tools ineffective at detecting AI Skill Bundle risks.
  • nyuwayskillscanner scans bundles before deployment to block high-risk installations.

How to Write a 1980s Photo Prompt That Looks Like a Real 1985 Snapshot

A good 1980s photo prompt doesn't just say "make it look retro." It tells the image model which camera took the picture, how the light hit the subject, and what was on the film.

  • Specify camera type (35mm film snapshot, disposable, Polaroid, or camcorder)
  • Mention film stock (Kodak Gold or Fujicolor) and light source (flash, fluorescent, neon)
  • Add period details (hairstyles, fashion, objects) without specifying exact year

More from Saturday 3 October →