About / Dario Cositore
I help teams put AI to work, and keep it working.
AI enablement and operations leader. Nine years in automation, training and operations, two of them building on frontier models. I evaluate the models, design the agentic workflows, set the governance, then train your team to run it without me. Based in Italy, working remotely worldwide.
The path here
No career pivot story. The same job, three levels deeper: make the work run, teach the people who do it, then hand the system back to them.
Sofia, BG
SBTech: trainer, then first B2B Learning & Development hire
Promoted from sales into training, then joined the B2B team as its first L&D hire. Rebuilt onboarding into a digital training library that cut new-hire ramp-up time 30%, and became the product-knowledge resource across three offices. First lesson in what enablement does to a business.
Remote
Proto AICX: Product & Process Specialist
First-generation AI chatbots for customer service. Built training frameworks, analyzed transcripts with NLP, pushed first-contact resolution up 20%. Learned where machines handle work well and where they quietly break trust.
Remote / US
Operations leadership: a 30-person team through automation
Hired as a manager. Promoted twice. Rebuilt intake, CRM, data and reporting from spreadsheets into an automated stack, taking revenue from $1.2M to $4.8M with qualified conversions +45% and profitability +25%. Trained staff hands-on and wrote the SOPs that made it stick, while running a 30-person distributed team on top of it.
Italy / Remote
Independent: AI enablement & agentic systems
Selected for a domain-expert panel benchmarking 13+ pre-release frontier models across 100+ business workflows, documenting hallucinations and structural failures that fed internal eval rubrics. Architected and shipped 5 production agentic systems: CRM intake, document review, memory and state, all on REST APIs, Git workflows and human approval gates. Gapfeed processes 189K+ buyer signals monthly. And I run the enablement side with client teams: needs assessments, access control and governance standards agreed with IT and InfoSec.
How I think
Three principles. Every system I build obeys all three or it does not ship.
I fix your process first
Automating a broken process just breaks things faster. Before any agent touches your work, I define how the work should flow. Then the system gets built on top of something solid.
You stay in control
Audit what your agent did, override any decision, expand its scope without writing code. Approval gates wherever you want them. Logs that show you everything.
You keep it when I leave
If it needs me forever, I built it wrong. Documentation, training and monitoring ship with every build. It runs without me. Most clients only call once a quarter.
What I build
Four ways I work with teams. One goal: your people using AI reliably, without me in the room.
Enablement
AI enablement & workflow design
Needs assessment and workflow discovery first, then agentic workflows built around how your team already works. Approval gates, permission boundaries and governance standards agreed with your technical and InfoSec teams before anything ships. See the services.
Reliability
Model evaluation & AI governance
Model selection, red-teaming, failure-mode analysis, eval rubrics and observability. I have documented hallucinations and structural failures on 13+ pre-release models, so you find out where the model breaks before your customers do. See the model evaluation case study.
Agents
Custom agents & document processing
Production agents with memory, permissions and approval flows. OCR, classification and extraction for contracts and forms. SaaS onboarding agent and lead qualification engine.
Want this running in your business?
Start with the free diagnostic: paste your broken workflow, get a written technical analysis within 48 hours. Or go straight to the rate card.
Stay in the loop.
Technical breakdowns, new system builds, and lessons from deploying autonomous AI. Direct to your inbox.
Weekly architectural updates. Unsubscribe anytime.