Building Agentic AI Foundations: How Intel® Liftoff Startups Are Preparing for the Next GPT Moment

Authors: Ed Lee, Kelli Belcher, Alex Sin - software solutions engineers at Intel and tech mentors for Intel Liftoff program

The rise of agentic AI is setting the stage for the next major leap in software. Intelligent agents can now plan, reason, and act autonomously across complex workflows. Gartner projects that by 2028, 33% of enterprise software will incorporate agentic AI.

This year’s Intel® Liftoff Days Q1 2025 event showcased just how far and fast early-stage teams can go when given the right tools, mentorship, and infrastructure. Held virtually over five days, Intel® Liftoff Days dev sprint helped eleven AI startups sprint toward technical milestones using Intel® Tiber™ AI Cloud, Gaudi® accelerators, and open source frameworks like OPEA.

Teams worked across a range of AI use cases from renewable energy and cell therapy to search-and-rescue drones. What they shared was a commitment to laying a strong foundation for the practical adoption of agentic AI everywhere that it’s needed.

Four Pillars of Agentic AI, in Action

At the heart of recent progress is a shift in how startups think about autonomy, orchestration, and system design. During Liftoff Days and recent workshops, startups focused on four critical areas to make agentic AI viable.

1. Designing Better Agents

From day one, Liftoff startups explored how much autonomy to give their AI systems and what toolsets were needed to support that agency. Startups like Pixel ML used their time to enhance AgenticFlow.ai, blending flexible tool orchestration (via Anthropic’s MCP) with structured workflows. This design allows agents to generate complex outputs like stories, visuals, and marketing assets, without going off track.

Meanwhile, workshops covered open-source agentic stacks including LangChain, LlamaIndex, and OPEA. giving teams practical foundations to create intelligent workflows that are both customizable and upgradeable.

2. Organizing Knowledge for Better Decisions

Autonomous agents are only as good as the knowledge they can retrieve. Several teams focused on improving knowledge representation to boost contextual accuracy in agentic execution.
Kneogin Igmisarch explored concept-retrieval frameworks like GraphRAG to better map domain-specific knowledge. Meanwhile, Qdrant demonstrated a hybrid system combining vector databases and graph structures, showing how LLM agents can pinpoint relevant data faster and with more nuance.

These approaches help agents move beyond keyword-based search and into true domain-aware reasoning.

3. Fine-Tuning for Real-World Tasks

Precision is critical in high-stakes environments like energy, medicine, and emergency response. That’s why startups like Reama AI and ParaWave focused on domain-specific fine-tuning. Reama used Liftoff Days to optimize their chatbot for renewable energy field ops and regulatory reporting, while ParaWave enhanced their drone AI to better support search-and-rescue missions.

Intel-led sessions on synthetic data generation and model-as-a-service (MaaS) also helped teams simulate real-world conditions, allowing them to fine-tune LLMs quickly without needing huge datasets upfront.

4. Making AI Explainable

Agentic AI isn’t just about execution. It’s about trust. Startups like AiCella, which is accelerating cancer cell therapy manufacturing, worked to make their systems auditable and transparent. That includes showing which sources were prioritized, how risk was assessed, and how conclusions were drawn.

Techniques like chain-of-thought and LLM-as-a-judge were explored to turn "black box" decisions into verifiable, human-readable steps. This leads to agentic workflows that regulators and domain experts can inspect, validate, and trust.

Intel® Liftoff Days: Where Real Progress Happens

The sprint format of Intel® Liftoff Days proved that with the right support, startups don’t need months to make a leap. Teams arrived with a diverse set of goals that included things like real-time video inference and multi-agent reasoning. They left with working demos, new insights, and access to extended cloud infrastructure.

A few highlights:

• Parawave integrated real-time drone video sharing with Intel Tiber™ AI Cloud, earning them free access to vision-language models.
• Reama AI added multi-agent features and accelerated chatbot performance, receiving beta access to LLM APIs.
• The Mango Jelly swiftly plugged Intel’s APIs into their shopping assistant, showing immediate generative commerce value.
• Pixel ML and Parawave began a promising collaboration exploring agentic reasoning in live drone footage: a real-world use case we’ll be sharing more about soon.
  
  

Looking Ahead

We’re already gearing up for the next  Intel® Liftoff Days, where new startups will join the community and explore what’s possible with Intel’s open AI stack.

If you’re an early-stage AI startup ready to scale your product, test your models, and join a global community of builders, explore the Intel® Liftoff for Startups program.