The State of AI Automation in 2025

This report analyzes 3,342 pieces of content from YouTube and Hacker News published throughout 2025, tracking mentions of automation platforms, LLM providers, architectural patterns, use cases, and integrations.

The data reveals a market in rapid transition: n8n has emerged as the dominant low-code platform, nearly 40% of automations now use agent-based architectures, and video generation automation became practical for the first time.

Key Findings

1. n8n Dominates the Low-Code AI Automation Market

Among content that explicitly mentions an automation platform, n8n captures 90% of mentions (1,387 out of 1,544 identified). Make's share dropped from 31% in Q1 to just 4% in Q4.

Quarterly breakdown (share of content where platform was identified):

• Q1: n8n held 69% vs Make's 31%
• Q4: n8n at 96% vs Make at 4%

This dominance in automation content appears driven by a combination of factors: n8n's open-source model and AI integrations (including LangChain-powered agent nodes), and its strong adoption by the YouTube creator/automation educator community.

2. The Rise of AI Agents: 40% of Automations Now Agentic

The automation paradigm is shifting from static workflows to intelligent agents. Of content where architecture was identifiable:

Nearly 40% of automations with identifiable architecture use single-agent patterns — AI systems that can autonomously decide which tools to use and iterate toward goals. Multi-agent systems (multiple AI agents collaborating) remain relatively rare at 2.5%, suggesting the coordination challenges aren't yet solved for most use cases.

RAG (Retrieval-Augmented Generation) accounts for 2.4% of identified architectures and tripled over the year (Q1: 6 → Q4: 18). Despite being a more established pattern, RAG remains relevant for grounding LLM outputs in verifiable sources — a market projected to grow 42% annually.

3. DeepSeek's Brief Moment

DeepSeek R1, released in January 2025, drew significant attention as an open-source reasoning model with performance comparable to GPT-4 at a reported training cost of ~$6M — a fraction of OpenAI's investment.

Our data shows the automation community took notice, with 37 mentions in Q1 (6.5% of LLM mentions in that quarter), but interest faded quickly:

Despite strong benchmark results, DeepSeek did not gain sustained traction in automation workflows. Practical integration into platforms like n8n requires API stability, documentation, and ecosystem support — areas where established providers have a significant head start.

4. Video Generation Becomes Automatable

2025 marked the year AI video generation tools became practical for automation:

The mention patterns track precisely with release dates. Sora launched publicly on December 9, 2024, yet automation content only exploded in Q4 2025 — coinciding with Sora 2's September 30, 2025 release. Veo's Q2 spike (0 → 17 mentions) aligns with Veo 3's announcement at Google I/O in May 2025, and interest remained steady as Veo 3.1 followed in October.

Together, these tools enable a capability that wasn't practical before 2025: automated video production pipelines.

5. WhatsApp Automation is Surging

The most dramatic integration growth story is WhatsApp:

WhatsApp became the #1 integration by volume (90 total mentions), surpassing traditional automation targets like Gmail (31) and Slack (19). This reflects the global shift toward messaging-first customer engagement — WhatsApp is the dominant messaging platform in many continents, with over 2 billion monthly active users. As customer engagement moves to messaging, automations are following.

6. Voice Synthesis: A Competitive Market

While video generation rose, ElevenLabs mentions in automation content decreased:

This 54% drop (Q1 → Q4) could reflect either market commoditization or ElevenLabs becoming so standard that creators no longer feel the need to mention it explicitly. Both interpretations have merit: OpenAI's TTS API lets developers add voice synthesis without a separate integration, while open source alternatives like Coqui TTS, IndexTTS (Feb 2025), and Muyan-TTS (Apr 2025) now rival commercial quality. At the same time, ElevenLabs may simply have become the assumed default for voice synthesis in automation workflows.

7. Google's Gemini Surge

Google LLM mentions grew 656% from Q1 to Q4, the most dramatic provider growth of 2025:

The Q4 spike aligns precisely with Google's release cadence:

• March 2025: Gemma 3 brought multimodal capabilities to Google's open-source line
• May 2025: Gemini 1.5 Flash became the go-to model for cost-sensitive automation (10x cheaper than GPT-4)
• November 2025: Gemini 3 launched with DeepThink reasoning capabilities

For context, OpenAI stayed flat (80 → 88 → 63 → 90) while Google climbed from distant third to a solid second position at 19.4% of identified LLM mentions. The combination of competitive pricing, strong n8n integration, and rapid model iteration made Gemini increasingly attractive for production automation workflows.

Use Cases vs. Content Types

We classified content into two categories: use cases (specific business functions being automated) and educational content (tutorials, architecture discussions, and meta-content about the automation industry). The split is nearly even — 49% use cases, 51% educational.

Use Cases: What's Actually Being Automated

Of the 1,632 items describing actual automation use cases we got the following distribution:

Content Production and Sales/Marketing together represent 60% of identifiable use cases. This might be enabled by the enhanced video generation models discussed above, and might potentially lead us to see more "AI content" on social media platforms in the future.

Content Types: What the Automation Community Creates

The remaining 1,710 items (51% of all content) are meta-content about automation rather than automation use cases:

Key insight: The automation content ecosystem is primarily educational. Platform training (42%) and agent architecture (22%) together account for two-thirds of non-use-case content — this might be due to the bias we have in our data by just pulling YouTube and Hacker News content.

The "Agency & Monetization" category (17.5%) reveals the emerging "AI automation agency" business model — content about selling automation services rather than just using them.

LLM Providers: OpenAI Still Leads

Most automation content (82%) doesn't explicitly mention which LLM powers the system. Of the 604 items that do we get the following chart:

OpenAI maintains a comfortable lead, while local models still remain niche.

Methodology

Data Sources

This report analyzes content from two sources:

1. YouTube — Videos from automation-focused channels. We identify these by searching for the most relevant channels that contain "AI Automation" in the name. We take the top 50 channels, and pull all their video titles and descriptions for 2025. This results in 2670 content pieces
2. Hacker News — Stories matching AI automation keywords that appeared in 2025, resulting in 672 content pieces.

This process was implemented using n8n and gave a total of 3342 content pieces.

Annotation Process

Each piece of content was classified across five dimensions using an LLM-based annotation pipeline:

1. Platform — The automation tool (n8n, Make, Zapier, etc.)
2. LLM Provider — The AI model provider, normalized (ChatGPT → OpenAI)
3. Architecture — The automation pattern (workflow, single_agent, multi_agent, rag)
4. Use Case and Content Type — The business domain or content type (content_production, sales_marketing, etc. or tutorial etc.)
5. Integrations — External services mentioned (WhatsApp, Gmail, etc.)

Annotation Guidelines

The annotation model was given explicit rules:

• Only label what is explicitly mentioned — no inference from context
• Normalize product names to providers (GPT-4, ChatGPT, o1 → "OpenAI")
• Default to "undefined" when classification is unclear
• For architecture: "workflow" is the default; "agent" requires explicit agent terminology

Limitations

1. Source bias — YouTube content may be skewed towards tutorials and promotional material as well as trendy topics such as n8n. In addition, we considered only the top-50 AI Automation channels.
2. High undefined rates The content title and descriptions are short, leading to high undefined rates — 54% of items have undefined platform, 82% have undefined LLM provider
3. Creator concentration — A small number of prolific YouTube channels produce disproportionate content
4. English-language focus — We consider content in English only.

Conclusions

Platform Consolidation

n8n's 90% share of identified platform mentions reflects its strategic positioning: open-source flexibility combined with native AI integrations. Make's drop from 31% to 4% quarterly share suggests limited room for competing low-code platforms in automation content.

Agent Adoption Patterns

Single-agent architectures now account for 40% of identified automation patterns, demonstrating production-ready maturity. Multi-agent systems remain at 2.5%, suggesting orchestration complexity continues to limit adoption despite theoretical benefits.

Video Generation Integration

The emergence of Sora and Veo as automatable tools marks a capability shift. Automated video production pipelines — previously impractical — are now viable for content workflows.

Messaging-First Customer Engagement

WhatsApp's position as the leading integration (90 mentions) over traditional targets like Gmail (31) reflects broader shifts in customer communication. Automation tooling is adapting to where engagement occurs.

Have questions or want to explore the data further? Contact us

Disclaimer: This report is an independent analysis based on publicly available metadata from YouTube and Hacker News. It does not reproduce or redistribute original content. All trademarks belong to their respective owners. Findings reflect trends in the analyzed dataset and are not endorsements or commercial recommendations.