Caddi Launches Agents From Screen Shares

// 1d agoPRODUCT LAUNCH

Caddi Launches Agents From Screen Shares

Caddi launched an agent that discovers repetitive back-office work, learns workflows through narrated screenshares in Loop Studio, and builds governed agents across existing business applications. Its hybrid architecture uses AI for judgment, deterministic code for exact execution, with scoped permissions and replayable logs. [Announcement](https://www.trycaddi.com/blog/caddi-launches-agent-that-builds-back-office-agents)

// ANALYSIS

Caddi’s strongest idea is starting automation with operational discovery and exception capture, rather than a blank workflow canvas. The real test is whether demonstrations remain reliable as policies, interfaces, and edge cases evolve.

–Discovers and ranks repetitive processes by frequency and cost, helping firms prioritize automation.
–Converts answers to edge-case questions into explicit operating rules.
–Uses AI selectively for document interpretation and classification while keeping repeatable steps deterministic.
–Targets law, wealth management, and insurance firms where auditability and permissions are essential.
–Its 100-plus integrations and SOC 2 Type II compliance make it more deployable than typical screen-recording automation. [Product overview](https://www.trycaddi.com/product)

// TAGS

caddiagentautomationno-codecomputer-usetool-useobservability

DISCOVERED

1d ago

2026-08-28

PUBLISHED

1d ago

2026-08-28

RELEVANCE

7/ 10

AUTHOR

[REDACTED]

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

RESEARCH1h ago

Anthropic’s AI Researchers Mitigate Alignment Failures

Anthropic’s Claude-powered agents autonomously propose and test post-training methods targeting ten alignment failures, including deception, sycophancy, jailbreaks, and prompt injection. The strongest methods generalized to held-out benchmarks, multi-turn audits, and models up to 4.7× larger while preserving measured capabilities. [Anthropic](https://alignment.anthropic.com/2026/automated-alignment-researchers/)

NEWS1h ago

OpenAI previews Astra with persistent agents

OpenAI previewed Astra, an unreleased model family that coordinates multiple agents, navigates desktop software, and supports persistent autonomous research. Sam Altman says OpenAI could have an internal system he considers AGI by year-end.

UPDATE1h ago

Tesana Loop automates game-building iterations

Tesana’s new Loop feature gives its AI game-building agent a token budget, then lets it build, test, and rebuild autonomously until the game is complete. The workflow extends Tesana’s existing code-writing and self-correction loop on Godot. [Tesana documentation](https://docs.tesana.ai/how-tesana-works)