Agents at work

In this episode of Agents at Work, Jordi Montes talks with Ben, the builder behind Polsia.

Ben is building an AI “founding team” that can spin up and run online businesses end-to-end.Instead of asking “what can agents do in theory?”,

He runs the growth loop for real: generate an idea, ship a landing page, wire up Stripe, run outreach/ads, respond to customers, ship fixes, and repeat without a human babysitting every step.

They get into:

- The autonomous growth loop: idea → build → launch → traffic → revenue → learn → iterate

- Why the winning move is speed + distribution, and why startups that aren’t “80% autonomous” get outpaced

- What it takes to give agents the full stack (web server, DB, GitHub, email, payments, ad accounts) so experiments run end-to-end

- Ben’s approach to shipping fast without breaking everything (model cross-checking + safer deploy decisions)

- How per-company memory + shared “best practices” turns one agent into a compounding advantage across many businesses

- The real business question: do you sell software, sell tokens, or take a cut of outcomes?

If you care about shipping faster than your competitors, turning experiments into compounding systems, and using agents to run the messy parts of a business (not just generate copycats) this one’s for you.

In this episode of Agents at Work, Jordi Montes sits down with Yunfan, an ex-Google / ex-Meta (Llama 3) engineer now building agents at Yutori, to talk about what it actually takes to ship agents that run continuously and stay reliable in the real world.

Yunfan breaks down Scouts, Yutori’s agentic web search product that can run on a schedule (daily/weekly/hourly), browse like a human, and only notify you when something meaningfully changes without spamming your inbox.

They explore:
What “agents” really are: a model + a loop of action → observation → next action (and why that definition matters)
Multi-agent orchestration: orchestrator + specialized agents (travel, finance, info gathering) and how “division of labor” improves performance
MCP vs function calling: why Yunfan thinks many tool integrations can collapse into simple functions/scripts ( “bash is all you need”) with less complexity and less context overhead 
Agents need memory: why Yutori is evolving from a report archive to a real “file system” so agents can store progress, preferences, and reusable knowledge 
The real problem: reliability: why 95% accuracy fails at daily cadence and the push toward 99–99.9% (plus supervision/self-healing)
Model economy: why not all tokens are equal, and how mixing frontier models with smaller open models makes agents affordable at scale
If you’re building agents that have to run long workflows, browse the web, and deliver trustworthy outputs over time, this episode is for you.

In this episode of Agents at Work, Jordi sits down with Gil Feig, co-founder of Merge, to talk about the unglamorous "plumbing" that suddenly matters even more in the age of AI: integrations and data access.

Gil breaks down why everyone having access to the same LLMs makes proprietary advantage shift toward who has the best data, the best access patterns, and the most reliable connections and how Merge is evolving from unified, synced integrations to agentic, MCP-style live calling.

They explore:
Sync vs live access: why copying + normalizing datasets helps retrieval, but comes with real cost and limits-
What MCP really is (and isn’t): “list tools” + “call tool” in practice
The security trap of tool-calling—and why Merge built DLP + approvals into the workflow
How Merge’s Agent Handler shows up in production (an MCP server backed by connectors + security)
Pricing in the agent era: per tool call for the new product, and per connected customer for the classic one

If you’re building agents that need to do real work and you want a sober take on what breaks in production, this one’s for you.

In this episode of Agents at Work, Jordi Montes sits down with Olga Beregovaya, VP of AI at Smartling, one of the leading enterprise translation and localization platforms. Together, they unpack the real state of multilingual AI and what it actually takes to translate at global scale.

They explore:
- The evolution of translation tech. From rule-based systems to statistical models to modern LLM-driven workflows
- Why enterprise translation is a completely different game than consumer tools
- The role of data cleanliness, linguistic assets, and centralization in delivering global content
- What “agentic translation pipelines” are and why they outperform vanilla LLM translation
- How hyper-localization, hallucination mitigation, and post-editing workflows are reshaping global content operations
- Why enterprises are rethinking “build vs buy” as LLMs become deceptively easy to prototype but hard to productionize
- The surprising limits of current transformer models and why purpose-built, smaller models may be the future

Olga shares 20+ years of experience in NLP, from rule-based MT to neural models to today’s multimodal systems. She explains how Smartling approaches accuracy, latency, brand voice, compliance, and global scalability for customers like Disney and IBM and why translation is becoming a business outcome, not just a linguistic task.

In this episode of Agents at Work, Jordi Montes sits down with Rodrigo Stevaux to explore how logic, formal methods, and AI are converging. Rodrigo is an economist turned technologist, researcher, and builder.

They discuss:
• How Rodrigo went from venture capital to deep tech and formal verification
• What “formal methods” really are and why proving correctness matters more than testing
• How logic programming (like Prolog) can make AI agents safer, smarter, and more deterministic
• The revival of symbolic reasoning and its link to modern “neuro-symbolic” AI
• Why knowledge bases and graph databases are secretly the same thing
• The missing link between today’s prompt-based agents and tomorrow’s reliable systems

Rodrigo shares his experience bringing old-school rigor to modern AI, from using state machines in agent design to mixing Prolog with LLMs for true reasoning. 

Together, they unpack why specification is the new code, and how the next breakthroughs in AI might come not from more data, but from better logic.

If you’ve ever wondered how we can make AI agents reason, not just predict this conversation is a must-listen.