How is agentic slide editing different from AI slide generation?

AI slide generation builds slides from a blank prompt. Agentic slide editing reads your existing slides, makes targeted changes to specific shapes, verifies the output, and iterates until the task is complete — without regenerating the slide or losing your existing formatting and data.

Which PowerPoint tools use agentic slide editing?

Deckary is the primary PowerPoint add-in with a built-in agentic editing loop — reading slides, issuing tool calls across 30+ edit actions, and verifying results through up to 15 iterations. Claude for PowerPoint (launched February 2026) uses a conversational editing approach inside PowerPoint but does not have Deckary's specialized consulting chart types.

What is the difference between agentic AI and regular AI in PowerPoint?

Regular AI in PowerPoint generates content on request and stops. Agentic AI operates in a loop: it senses the current state of your slide, reasons about what to change, takes action through tool calls, observes the result, and repeats until the task is done — similar to how agentic coding agents read code, write changes, run tests, and fix errors iteratively.

Can agentic slide editing preserve existing slide formatting?

Yes. Because agentic editing reads and modifies specific shapes rather than regenerating slides from scratch, it preserves existing fonts, colors, chart data, and layout. This is the core advantage over generation-only tools, which typically overwrite or lose existing formatting when they rebuild a slide.

Agentic Slide Editing: The New AI Paradigm for PowerPoint

Name: Deckary
Rating: 4.8 (50 reviews)
Author: Deckary

Agentic slide editing is AI that reads, edits, and verifies existing slides through a tool loop — not just generation. Here's how it works and why it matters.

Bob · Former McKinsey and Deloitte consultant with 6 years of experienceApril 3, 20268 min read

Pricing and feature information was accurate at the time of publication. Competitor products change frequently — verify current details on each provider's website.

Agentic slide editing is a new AI paradigm for PowerPoint: instead of generating slides from scratch, the AI reads your existing slides, issues targeted edits through a tool loop, verifies the result, and iterates — the same way agentic coding agents work with code files. After analyzing Deckary's implementation and the broader field of AI presentation tools, we believe this distinction matters more than most teams realize.

The difference is architectural. Generation builds something new. Agentic editing works with what exists.

Agentic slide editing loop diagram — read, analyze, edit, verify, iterate

What Is Agentic Slide Editing#

Agentic slide editing is the application of an agentic AI loop to existing PowerPoint slides. The loop has four stages:

Read — the AI reads the actual slide content: shape names, text, positions, chart data, and the slide master
Analyze — the AI reasons about what needs to change relative to the stated objective
Edit — the AI issues targeted tool calls to specific shapes (edit text in shape X, move shape Y, resize shape Z)
Verify — the AI takes a snapshot of the result and checks whether the edit was successful; if not, it retries

This loop runs repeatedly — up to 15 iterations in Deckary's implementation — until the task is complete or the iteration limit is reached.

The term "agentic" comes from the broader AI field. IBM describes agentic AI as systems that operate through a four-step "perceive, reason, act, learn" loop — completing goals with minimal human supervision. Google Cloud defines agents as applications that "achieve a goal by processing input, performing reasoning with available tools, and taking actions based on its decisions." Agentic slide editing applies this loop to the specific environment of a PowerPoint file.

How It Differs from AI Slide Generation#

AI slide generation and agentic slide editing are often conflated. They address different problems.

Capability	AI Slide Generation	Agentic Slide Editing
Starting point	Blank canvas or prompt	Your existing slides
Reads current content	No	Yes
Preserves formatting	No (rebuilds from scratch)	Yes (edits specific shapes)
Verifies result	No	Yes (snapshot after each tool call)
Iterates to fix errors	No	Yes (up to 15 iterations)
Use case	First draft, new sections	Revisions, data updates, restructuring

The analogy to software development is direct. Code generation (GitHub Copilot autocomplete, ChatGPT code) produces new code and stops. Agentic coding (Copilot Workspace, Cursor agent, Claude Code) reads the existing codebase, writes a change, runs tests, reads the error output, and fixes the problem — iterating until the tests pass. Agentic slide editing is the presentation equivalent.

PPTAgent, a research framework published at EMNLP 2025 (arxiv.org/abs/2501.03936) by researchers at the Institute of Software, Chinese Academy of Sciences, formalized this distinction. The paper describes a "beyond text-to-slides" approach where an AI agent issues editing actions on reference slides rather than generating from a blank canvas — demonstrating that the research community has identified agentic editing as a fundamentally different capability from generation.

Continue reading: Bar Charts in PowerPoint · Best Fonts for PowerPoint · 30-60-90 Day Plan Template

Generate consulting slides with AI

Describe what you need. AI generates structured, polished slides — charts and visuals included.

Try AI Slide Builder See All Features

Why Agentic Editing Matters for Professional Presentations#

Empower Suite's "Big PowerPoint Study" (conducted by Nielsen) found that 37% of time working with PowerPoint is spent on formatting — not on analysis, writing, or design decisions, but on execution: moving shapes, adjusting font sizes, fixing alignment after content changes.

AI slide generation doesn't address this. It produces a first draft, but the formatting work happens after generation — and if the content changes (the data gets updated, the recommendation shifts, the client requests a different layout), the generation needs to run again, and the formatting gets overwritten.

Agentic editing addresses the revision cycle directly. When a partner sends revised Q4 numbers the night before a board presentation, the agentic loop reads the existing charts, updates the data in the relevant shapes, verifies the result, and preserves everything else — the fonts, the spacing, the labels, the surrounding slide content.

For consulting and investment banking workflows where presentations go through 5-15 revision rounds before client delivery, this is a structural improvement over generation-only AI tools.

The Tool Loop: How Deckary's Implementation Works#

Deckary's agentic editing loop, accessible from the AI pane inside PowerPoint, operates through 30+ tool calls that map to specific PowerPoint actions:

Read tools:

read_slide — returns all shapes, text, and positions on the current slide
read_all_slides — returns the full presentation structure for cross-slide reasoning

Edit tools:

edit_text — modifies the text content of a specific named shape
format_text — adjusts font size, weight, or color within a shape
move_shape — repositions a shape by name to specific coordinates
resize_shape — changes the dimensions of a shape

Verification: After each write tool executes, the loop automatically injects a verification snapshot — a read of the modified shape — so the AI can confirm the edit applied correctly before issuing the next tool call. This is what distinguishes the loop from simple command execution: the AI observes the result of its own actions and can correct them.

The loop runs up to 15 iterations per task. Complex restructuring tasks (reformatting a multi-section slide, updating all chart titles across a deck, standardizing bullet formatting) typically complete in 3-8 iterations. The user sees each tool call in the AI pane as it executes.

Which Tools Offer Agentic Slide Editing#

As of April 2026, two PowerPoint add-ins have meaningful agentic editing capabilities:

Deckary — a PowerPoint add-in ($49-119/year) with a built-in agentic editing loop, 30+ tool calls, up to 15 iterations per task, and verification snapshots after write operations. Also includes consulting-specific capabilities: waterfall and Mekko chart generation, 2,000+ icon library, and AI slide builder for generating new slides. Works on both Windows and Mac.

Claude for PowerPoint — launched February 2026 as a Microsoft 365 add-in (included with Claude Pro at $20/month and above). Reads the slide master and brand guidelines to generate on-brand content. Supports conversational editing of existing slides and "Agent Skills" — reusable workflows saved inside the sidebar. Strongest for template fidelity; does not include specialized chart types like waterfall or Mekko.

Standalone tools (Gamma, Beautiful.ai, Canva) and non-agentic AI add-ins (standard SlidesAI, earlier Plus AI builds) do not implement an edit-verify loop — they generate or restyle content without reading and modifying specific shapes on existing slides.

When to Use Agentic Editing vs. Generation#

The right tool depends on where in the workflow you are.

Use AI generation when:

You're starting a new deck from a brief or outline
You need a complete first draft in under 5 minutes
The slide doesn't exist yet

Use agentic editing when:

A presentation exists and needs revisions
Data has changed and charts need updating
A slide's structure needs restructuring without losing existing content
You need to standardize formatting across 10+ slides
A partner has marked up a slide and you're implementing the changes

The optimal workflow combines both: use AI generation for the first draft of new sections, then use agentic editing for revision cycles. Deckary is the only inside-PowerPoint tool that supports both in a single add-in — slide builder for generation, agentic editing for revisions.

Implications for AI Presentation Tools#

Most AI presentation tools today compete on generation quality: how good is the first draft? Agentic editing shifts the competition to revision quality: how well does the AI work with what already exists?

This distinction matters because in professional contexts, presentations are rarely built from scratch in a single session. Strategy decks go through partner review. Financial models get updated quarterly. Investor presentations evolve across multiple meetings. The generation phase is one hour; the revision phase is twenty.

The AI tools that will win in professional workflows are those that can edit as fluently as they generate. Agentic slide editing — reading existing content, reasoning about changes, executing targeted tool calls, verifying results — is the architecture that makes that possible.