AI Agents Need Deterministic Building Blocks
Techstrong TV Interviews
•
26m
In this Techstrong TV interview, Mike Vizard talks with Adam Jacob, co-founder and CEO of Swamp Club. They discuss why deterministic AI agents are the key to moving AI from prototype to production. Swamp Club has raised $7 million in seed funding. Its bet is that the next phase of AI adoption will reward repeatability and cost control, not just bigger models.
Why AI Agent Prototypes Fail to Repeat
Most teams quickly discover that AI can do almost anything once. However, getting it to do the same thing the same way the next time is much harder. Swamp addresses that gap by giving agents primitives. These are modular components with typed inputs and outputs that store their data and behave predictably. As a result, frontier models handle only the moments that require judgment, while everything else runs as reliable code.
Deterministic AI Agents Cut Token Costs
Jacob describes a contractor onboarding process that once required five teams to coordinate across eight systems. Altogether, it took six to eight weeks. Rebuilt with Swamp, roughly 99% of that agent is now deterministic code. In another case, a code review workflow used eight times fewer tokens. By contrast, agents loaded with skills and MCP connections often improvise every step. Jacob calls them "the most expensive for loops that have ever existed." In his view, treating AI as an all-knowing machine is the wrong mental model.
Software Factories, Deterministic Agents and ROI
The conversation also covers software factories. In this model, organizations encode their architecture, accessibility and testing standards into workflows. Deterministic AI agents then follow them every time. Asked whether AI faces an ROI backlash, Jacob says that is the outcome he is most nervous about. Still, he believes the industry can make the numbers work. The key is for engineers to build practical, repeatable systems quickly.
Personalized Applications Built by Agents
Looking ahead, Jacob expects agents to make highly personalized applications practical. For example, a business partner of his built a financial projections application that pulls directly from NetSuite and the company's CRM. Previously, that kind of tool would have required a dedicated development project.
For more information, visit swamp.club.
Watch Mike Vizard's earlier conversation with Adam Jacob, Rethinking DevOps and Infrastructure as Code with System Initiative's Adam Jacob, and explore more Techstrong TV interviews.
Up Next in Techstrong TV Interviews
-
OpenAEV Brings Attack Chaining to Sec...
Annual pentests can't show how attackers chain small weaknesses into a full breach. Alan Shimel talks with Sebastien Miguel, senior product manager of OpenAEV at Filigran, about attack chaining, adversary exposure validation and how teams can test their defenses the way real attackers operate.
...
-
What's Blocking Fully Autonomous Vibe...
AI can generate code faster than ever, but running that code in production is still the hard part. Alan Shimel talks with Assaf Sauer, founder and CTO of Stacktic.io, about what is blocking fully autonomous vibe coding and operations, and how Stacktic aims to close the gap.
From presales arc...
-
Your AI Agents Will Find the Identity...
AI agents don't need to break your identity controls; they just find the shortcuts. Alan Shimel talks with Roy Katmor, co-founder and CEO of Orchid Security, about identity dark matter, why agents escalate access so quickly and how enterprises can get ready for AI.
What identity dark matter ...