Concept Graph, Resume & KeyIdeas using Moonshot Kimi K2 0905:

Resume:

Maximiliano Flores, CTO of InConcert, explains that agentic AI means systems able to pursue complete business goals, not just answer isolated questions. They decompose a user request into sub-tasks, choose tools, query databases, call APIs and decide autonomously while remaining auditable. InConcert’s year-old platform InAsian orchestrates fleets of such agents that share knowledge and hand conversations across channels, languages and tones without losing context, giving clients a gradual path from 100 % human to 100 % automated service while preserving a human safety net that “raises a hand” when confidence falls below a threshold. The architecture combines generative models with private knowledge bases distilled from years of client documents and videos, then wraps everything in fine-grained permission layers, output filters and real-time human supervision to avoid hallucinations, toxic language or data leakage. Flores insists this is not a laboratory proof of concept; dozens of companies already run revenue-producing workflows on the stack and the firm dog-foods the same agents internally for training and presales support.
Looking back, Flores’ first encounter with AI was at university in Uruguay in the early 2000s, but the real inflection came around 2010 when InConcert began using machine-learning for intent recognition in early chatbots. The current wave of generative models since 2022 has moved AI from cost-optimisation to strategic disruption: intelligence is becoming a commodity comparable to electricity, forcing every business model to rethink how value is created. Yet the speaker warns that 80 % of Spanish AI projects stall because companies buy the hype without defining measurable KPIs or change-management plans; success requires consultative selling that ties technology to end-to-end processes, measurable ROI and staged adoption. Regulation, ethics, cybersecurity and explainability must be engineered from day one, not bolted on later, because agents that act on behalf of a brand can amplify bias, leak secrets or be hijacked through prompt injection or back-doors. InConcert therefore invests heavily in red-team testing, audit logs and human-in-the-loop overrides.
Both guests agree society is entering an irreversible hybrid era where humans plus AI out-compete humans alone; entire professions will disappear or be redefined within five years, not decades, creating extreme inequality between AI-first and AI-absent organisations. The conversation closes by stressing that although artificial general intelligence may arrive by 2030, the immediate opportunity is practical: companies that build multidisciplinary teams, treat data as capital and iterate fast can already solve problems that were impossible twelve months ago. The invitation is to stop waiting for maturity and start experimenting today, because the technology, while still evolving, is already good enough to deliver differentiated customer experience, new revenue streams and sustainable competitive advantage if deployed with clear governance and ethical guardrails.

Key Ideas:

1.- Agentic AI autonomously decomposes complex goals into executable sub-tasks using external tools and data.

2.- InAsian platform orchestrates multi-agent teams that share knowledge and maintain context across channels.

3.- Human-in-the-loop “raise hand” mechanism intervenes when confidence drops, preventing brand damage.

4.- Private knowledge bases distilled from client documents reduce hallucinations and anchor responses in facts.

5.- Fine-grained permissions and output filters block sensitive data leakage during conversations.

6.- Real-time supervision console lets human supervisors correct agents and retrain models on the fly.

7.- Revenue-producing deployments already replace traditional contact-centre workflows in multiple industries.

8.- Internal dogfooding uses identical agents for employee training and presales support inside InConcert.

9.- Measurable KPIs and staged adoption plans are mandatory; 80 % of Spanish AI projects fail without them.

10.- Intelligence is becoming a commodity; business models must reinvent value creation beyond cost cuts.

11.- Regulation, ethics and security must be designed from day one, not retrofitted after deployment.

12.- Prompt-injection and back-door risks require continuous red-team testing and audit trails.

13.- Multidisciplinary teams combining business, legal and technical skills outperform pure tech groups.

14.- Five-year horizon foresees mass profession displacement, demanding rapid workforce reskilling strategies.

15.- AI-first companies already achieve higher revenue multiples than peers relying on traditional software.

Interviews by Plácido Doménech Espí & Guests - Knowledge Vault built byDavid Vivancos 2025