FULL AGENDA
3 Days · 6 Tracks
4 Simultaneous Venues
Six stages running in parallel. Choose your track. Each column is a different world.
● C-Level Forum● AI&TECH● SMEs● Industry● Impact● International
Six stages running in parallel. Choose your track. Each column is a different world.
Attendee check-in in the gardens of the Palacio de la Magdalena. Coffee with views of the Cantabrian Sea. The first visual impact only Santander can deliver.
Institutional opening ceremony of FoundAItion 2027. Words from the Government of Cantabria, the Mayor of Santander, the Rector of the UC and the Founder of AIRES.
The harness —tools, memory, the loop, verification and guardrails around the model— now matters as much as the model itself. A good harness makes a mid-tier model production-ready; a bad one makes even a frontier model unreliable. From prompts to harnesses in four years.
The frontier born in 2026: an orchestration layer above the harness. The system starts itself on a timer, spawns assistants when needed and supplies work to itself, keeping many agents running continuously. Prompt → Context → Harness → Loop.
88% of organisations reported a confirmed or suspected AI agent security incident over the past year, and prompt injection remains the leading cause of production failures according to OWASP. In June 2026 Microsoft warned that poisoned MCP tool descriptions can make an agent leak data, with up to 200,000 exposed MCP instances counted in the wild. The five attack vectors and how to defend: Zero Trust, runtime guardrails and human-in-the-loop circuit breakers.
July 2026 packed an unprecedented release cycle —GPT-5.6, Claude Opus 5, Grok 4.5, Qwen3.7— and the distance between a closed frontier model and the best open-weight one is now measured in weeks, not years. What that pace means for your stack, your vendor contract and your independence: how to design so you can swap models without rebuilding the product.
The average enterprise AI budget went from $1.2M in 2024 to $7M in 2026 — and 73% still overshoot their projections. Inference alone now takes 85% of the spend. CDOs managing billions in cloud share how they regain control: tiered routing, per-use-case budgets and automatic cut-offs.
On 16 June 2026 the European Parliament approved the Digital Omnibus amendments, deferring most high-risk obligations to December 2027 and August 2028. What did land on 2 August 2026: the Article 50 transparency duties —disclosing that you are talking to a chatbot, marking AI-generated content and labelling deepfakes— and oversight of general-purpose models. What you must do now and what can wait.
Three SME CEOs and CEOE-CEPYME representatives share real cases. Budgets between 10K and 100K. What works, what does not and how to measure it.
The real state of AI in Spanish manufacturing. Where it is deployed, what ROI it generates and what barriers slow adoption in the industrial sector.
Academics and NGOs challenge the narrative of frictionless progress and demand transparency on the social consequences of enterprise AI.
Research from Microsoft and Stanford's Digital Economy Lab puts agentic tasks at roughly 1,000x the tokens of a standard chat exchange, while Gartner measures 5–30x more tokens per task. A multi-step agent that plans, retrieves, calls tools and self-corrects costs $0.10–$1.00 per completed task. What autonomous agents really cost — and how to budget for them.
The opening keynote of FoundAItion 2027. The AI Act as catalyst or brake on European innovation. Digital sovereignty and competitive advantage in the new regulatory framework.
A DeepMind researcher presents the real state of the art: what the frontier labs know that companies will take 2 years to understand.
An 8B open model fine-tuned on your domain beats a generic frontier model on narrow tasks, at a fraction of the cost and without a single record leaving your network. When tuning pays off and when it does not, what LoRA buys you versus distilling from the large model, how many examples you actually need, and how to evaluate before going to production.
Open weights does not mean open source, and the difference lands in your contract. Apache 2.0, MIT, community licences with user thresholds, acceptable-use clauses and restrictions on generated data: what each one lets you do in a commercial product, which attribution obligations you inherit, and the traps your lawyer finds too late.
Lunch for all attendees. Informal networking between sessions. The Startup Fair open.
Gartner estimates 40% of enterprise applications will have embedded agents by the end of 2026, up from less than 5% in 2025 — yet only one in five companies has a mature governance model for them. The CDOs of three IBEX heavyweights share their production architectures: what scales, what does not, and where they draw the line.
Enterprise API pricing fell 67% year-over-year, from $18.40 to $6.07 per million tokens. And the bill still goes up, because inference now accounts for 85% of the AI budget. The paradox of paying less per token and spending more every month — and what it means for your architecture, roadmap and budget.
Mercadona in predictive logistics, Mapfre in automated underwriting, CaixaBank in AI customer service. Real cases with a tight budget.
The major energy companies present their AI cases. Predictive maintenance, digital twins and power grid optimization at real scale.
The most uncomfortable track of the summit. Academics and regulators question the social impact of AI and demand transparency on its consequences.
Ramp's Series F valuation and metrics, cost-efficiency gaps from tiered model routing and the 38% open-source/open-weight adoption share. Real frameworks for AI ROI in 2027.
The 2026 data dismantles the narrative: neither the Anthropic study, nor the IMF analysis of Denmark, nor the Stanford AI Index finds an aggregate unemployment signal. What is visible is the funnel closing from below —employment for 22-to-25-year-olds falling in the most exposed occupations— and 35% of entry-level postings already demanding AI skills. The distinction that explains it: automation replaces, augmentation does not. Real salaries, retention and how Europe competes.
Hands-on red-teaming workshop on production systems, working through the real 2026 cases: poisoned MCP tools, exfiltration triggered by seemingly benign instructions and supply-chain attacks on packages and skill marketplaces. Prompt injection tops the OWASP ranking of agentic failures and attacks grew 340% year-over-year. Live demos.
Live demo of the most powerful no-code tools: Zapier AI, Make, n8n and Microsoft Copilot compared. Deployments in under a week.
Three infrastructure companies share how they deployed digital twins on real timelines. Architectures, costs and measurable results.
The University of Cantabria, UIMP and representatives from the Ministry of Education debate the future of education with AI. Personalization, equity and pedagogical sovereignty.
OVHcloud financials and restructuring, the AWS European Sovereign Cloud isolation model and Microsoft France testimony on data-access guarantees. Real alternatives to US hyperscalers. In English.
The first grand evening of FoundAItion 2027. Signature northern cuisine. DJ set. Terrace overlooking the bay. Speakers and VIPs.
Register now and choose your track. Early Bird active until January 2027.