Boutique technology studio, passionate people driving tomorrow's technology today, and then some ;-) (@conflict not for sale or transfers, dont ask)
San José, Costa Rica
Joined October 2012
- Tweets104
- Following77
- Followers131
- Likes31
AI agents don’t need more access. They need better boundaries.
Least privilege matters more than ever.
conflict.blog/blog/agents-ne…
#AIAgents #Cybersecurity #AI
CONFLICT retweeted
For defense AI, the perimeter isn’t a constraint.
It’s the whole point.
No external APIs. No data leaving the enclave. Local inference and governance.
What air-gapped AI actually requires:
calliopeai.blog/air-gapped-a…
The chat window is the easy part.
Permissions, approvals, and audit trails are where production gets real.
We open-sourced the boilerplate:
conflict.blog/blog/ship-the-…
CONFLICT retweeted
open-sourced a Django + Next.js boilerplate today with a @CopilotKit copilot wired in from first boot: forms + workflow engine, auth, permissions, Docker, the works.
The neat part is where the agent runs. inside the stack, not bolted on beside it. same session auth as every user, re-checks permissions on every tool call, records by external ID only. anything irreversible pops an approve/deny card.
net result: the agent can only do what the logged-in user can do.
Pydantic AI serving AG-UI from Django, Claude underneath. API key never leaves the backend (of course).
clone it, ./run.sh, add a key, build. MIT licensed.
github.com/ConflictHQ/boiler…
CONFLICT retweeted
Sovereign AI is not just about where data lives.
For EU government and defense, the bigger question is whether AI can run inside controlled infrastructure with governance, auditability and no critical external dependencies.
calliopeai.blog/sovereign-ai…
#SovereignAI
AI vendor lock-in is back, but now it hides in prompts, model behavior, evals, retrieval layers and workflows.
The question isn’t “Can we switch vendors?”
It’s “What will switching actually cost?”
conflict.blog/blog/vendor-lo…
#AI #EnterpriseAI
Vendor lock-in is back. It just looks different in the AI era.
Switching models is easy. Switching systems, prompts, evals, and workflows is not.
Read more: conflict.blog/blog/vendor-lo…
CONFLICT retweeted
For European banks, AI procurement is no longer just about performance.
DORA, GDPR, and the EU AI Act make sovereignty, resilience, and data control essential.
Read more: calliopeai.blog/sovereign-ai…
#SovereignAI #DORA #BankingAI
CONFLICT retweeted
Sovereign AI is not just a model or a cloud region.
It is architecture: data, runtime, governance, auditability, and control.
Mistral matters, but real sovereignty needs private AI workbenches.
More from Calliope:
calliopeai.blog/sovereign-ai…
DORA metrics were built for human delivery loops.
Agentic software work changes the signal.
When agents write, rerun, and revise code, teams need new ways to measure quality, drift, review, and risk.
Read more: conflict.blog/blog/dora-metr…
#AI #DevOps #AgenticAI
CONFLICT retweeted
Sovereignty is no longer just a compliance issue — it’s becoming an AI infrastructure design constraint.
Build for portability, policy, and control from day one.
Read more: calliopeai.blog/when-soverei…
#AI #DataSovereignty #EnterpriseAI
MCP is powerful, but it is not a settled standard yet.
Our take on how teams should think about it in production:
conflict.blog/blog/mcp-is-no…
#AI #MCP #AgenticAI
CONFLICT retweeted
AI agents only work in production when tool use is safe, observable, and recoverable.
Good read from Kahi:
calliopeai.blog/function-cal…
#AI #AgenticAI
Six-month roadmaps are outdated.
Planning still matters. Roadmap fiction does not.
Outcomes are durable. Features are negotiable.
conflict.blog/blog/why-we-ki…
#ProductManagement
AI made code cheaper to write.
But now the expensive part may be reviewing it.
That shift changes how teams build software.
conflict.blog/blog/cost-of-c…
CONFLICT retweeted
Hear me out: enterprise AI does not need more tools. It needs an operating system. Every enterprise has to get through the same 5 stages with their apps and agents, and the surface where AI meets the enterprise is converging fast. BROCS: Build, Run, Observe, Control, Secure.
CONFLICT retweeted
Apps and agents run across cloud accounts, on-prem datacenters, and edge environments. That is the OS underneath the AI surface. Own your data. Own your cloud. Simplify how you operate inside both.
leomata.com/blog/brocs
CONFLICT retweeted
Been in enterprise sales long enough to know: the best security product in the world loses to "but my team already uses ChatGPT on their personal laptops."
You don't win by blocking. You win by building something they'd rather use. @calliopelabs
CONFLICT retweeted
Coincidentally, also how videos are made: lots of pictures in sequence. Turns out the hard part isn’t pressing the button, it’s knowing what to keep, what to cut, and what’s worth shipping.
