Introduction: Automate without writing code
Small and medium-sized enterprises (SMEs) are under growing pressure to speed up processes, reduce manual errors, and compete with digital giants. Many business leaders are asking themselves:How can they automate critical workflows without becoming software engineers?The answer today lies inAI self-running workflowsWhat are AI self-running workflows for SMEs?
- Customer onboarding:Data capture, verification, and automatic entry into the CRM.
- Invoice management:Extraction of invoice numbers, amounts, and dates from PDFs, with automatic accounting entries.
- Content development:Creation of blog posts, emails, and product descriptions based on templates and market research.
- Operational reporting:Data collection from multiple sources, semantic SQL generation, and dashboard creation.
Key technologies powering this autonomy include LLMs for natural language interpretation, agents for task execution, and new libraries that connect agents to existing systems.
Three 2026 innovations redefining autonomy
The landscape has changed dramatically in recent months. Three releases are giving SMEs capabilities previously available only to large companies:1. NVIDIA Auto Ontology (Apache 2.0)
Auto Ontology is an open-source library that automatically generates a semantic layer from corporate metadata. By connecting agents to a catalog of tables, columns, and relationships, Auto Ontology ensures that SQL commands generated by agents are certified and secure.# Example of using Auto Ontology to map a prompt to SQL
from nvidia import auto_ontology
ontology = auto_ontology.load('sales_db_metadata.json')
question = "Show total sales by region in the last quarter"
query = ontology.generate_sql(question)
print(query)
# Output:
# SELECT region, SUM(amount) AS total_sales
# FROM sales.transactions
# WHERE transaction_date >= '2026-07-01'
# GROUP BY region;2. The Letter Company’s Letterpress-1
# Example prompt for Letterpress-1
def generate_growth_script(product_research, market_trends):
agent = LetterpressAgent()
script = agent.generate(
goal='Increase conversion rate by 15% in 90 days',
context=product_research,
constraints=market_trends
)
return script
print(generate_growth_script(research_data, trends_data))3. OpenAI Pando (Git backend)
Pando is a Git-compatible backend that reduces median push latency by about 80%. For self-running workflows, this means that whenever an agent creates or modifies a file, versioning and deployment occur almost instantly, keeping the repository up to date.End-to-end self-running workflow: A practical example of customer onboarding
Imagine an SME wants to automate the onboarding of a new customer. Here’s how an AI self-running workflow could be structured using the technologies mentioned above.Step 1: Data acquisition
- A web form collects the company name, email, and an initial CSV file of contacts.
- An ingestion agent (based on an LLM) validates the fields, extracts the company name, and stores it in a temporary cloud bucket.
Step 2: Information extraction with an LLM
- A prompt is sent to an LLM to extract key entities: industry, size, region.
- The agent writes the extracted JSON to a temporary data store.
Step 3: Semantic SQL generation with Auto Ontology
- Auto Ontology reads the CRM database schema and translates a natural language request ("Insert these details into the CRM") into a secure SQL instruction.
- The agent executes the query, inserting records into thecustomerstable and setting the status to "onboarded."
Step 4: Integration script creation with Letterpress-1
- Letterpress-1 analyzes the repository of existing integration scripts and market trends.
- It generates a Python script that sends a personalized welcome email, creates an API key, and assigns an initial project to the new customer.
- The script is automatically versioned with Pando, ensuring an immediate push to the staging branch.
Step 5: Validation and reporting
- A monitoring agent checks that the email has been sent, that the API key is active, and that the project is accessible.
- Aggregated data is written to a business intelligence dashboard, updated in real time.
Benefits for SMEs
- Reduced onboarding time:From days to a few minutes.
- Minimized errors:SQL commands validated by Auto Ontology eliminate data entry errors.
- Scalability:The same workflow can be replicated for hundreds of new customers without proportional development costs.
- Data-driven iteration:Optimizations generated by Letterpress-1 based on market trends keep campaigns competitive.
Checklist for getting started
If you are an entrepreneur or IT manager, follow these concrete steps to introduce self-running automation into your business:- Identify a high-impact use case:Choose a repetitive process (e.g., customer onboarding, invoice management).
- Assess your resources:Check the compatibility of your existing technology stack with Auto Ontology, Pando, and an LLM agent.
- Develop a prototype:Use an open-source LLM and a simple agent to create an MVP of the workflow.
- Implement a feedback loop:Collect performance data, adjust prompts, and measure ROI.
- Scale gradually:Add other agents (e.g., reporting, email marketing) as the first cycle stabilizes.
Conclusion: Why AI self-running workflows are crucial today
Useful resources
- Documentation for NVIDIA Auto Ontology:github.com/nvidia/autontology
- Letterpress-1 Beta:letter.company/letterpress-1
- OpenAI Pando:openai.com/pando