Weber Shandwick is one of the largest public relations and communications firms in the world, with strategists continuously synthesizing social signals, news coverage, competitive intelligence, audience sentiment, and cultural trends into recommendations for major global brands.

That synthesis work was valuable, but much of the operational process behind it remained manual. Analysts were spending significant time gathering information across fragmented systems, compiling research, and transforming raw data into structured strategic insight. As client expectations around speed and responsiveness increased, those workflows became increasingly difficult to scale.

Focused was brought in to architect and build a production-grade agentic AI platform that could automate large portions of that research process while still keeping strategists firmly in control of final outputs and decision-making.

The goal was not to replace Weber’s analysts. It was to give them systems that could handle research orchestration, synthesis, and information retrieval at machine speed so teams could spend more time on strategy, interpretation, and client counsel.

Research and Discovery

Before building the platform, Focused worked closely with Weber’s teams to understand how research and strategic analysis actually happened day to day.

The engagement centered around several high-value workflows where the gap between raw data and actionable insight was the widest:

• Identifying emerging cultural moments where brands could authentically participate in online conversations
• Mapping how different political or demographic audiences might respond to messaging
• Compiling competitive and acquisition intelligence for clients navigating mergers, acquisitions, or major market shifts
• Monitoring media narratives and rapidly synthesizing changes in public sentiment

During discovery, it became clear the challenge was not data access. Weber already had deep integrations with platforms like Glystn and PeakMetrics that aggregated large volumes of social, news, and paywalled content.

The bottleneck was the layer between the data and the strategist.

Analysts were manually coordinating research steps, refining searches, compiling outputs, validating sources, and synthesizing findings across multiple systems. In practice, highly skilled strategists were spending large portions of their day acting as human orchestration layers between disconnected tools and datasets.

Focused’s role was to build agent systems that could take over much of that coordination and synthesis work while still allowing humans to guide, review, and refine outputs throughout the process.

Solution

Focused designed and built a suite of production AI agents using LangGraph, LangSmith, and Google Vertex AI infrastructure.

The agents were surfaced through Weber’s internal AI platform, Halo, which operates on Google Agentspace and Vertex AI. One of the major architectural challenges of the engagement was integrating LangGraph’s orchestration model into an environment that had meaningful constraints around state management and human-in-the-loop workflows.

To solve that, Focused architected a proxy orchestration layer that allowed LangGraph agents to communicate cleanly with Weber’s Google infrastructure while preserving a unified interface for employees inside Halo. This abstraction layer enabled Weber’s teams to interact with agents through a consistent internal experience without exposing the complexity of the underlying orchestration systems.

The platform included multiple specialized agents designed around Weber’s highest-value workflows.

Cultural Opportunity Spotter

Focused built an agent that automated a previously manual daily workflow where analysts reviewed large volumes of vendor data looking for culturally relevant opportunities for brands to engage in emerging conversations.

The agent continuously analyzed social and web signals, identified meaningful trends, and generated structured daily briefs for strategists to review and refine.

Deep Research Agent

Focused also developed a multi-agent deep research system capable of breaking large strategic prompts into smaller research tasks distributed across parallel sub-agents.

Using LangGraph orchestration patterns, the system could:

• Decompose strategic questions into discrete research objectives
• Fan tasks out across multiple specialized retrieval agents
• Pull information from curated internal and external sources
• Maintain state across long-running workflows
• Synthesize findings into structured strategic reports

This architecture allowed Weber’s teams to conduct significantly more comprehensive research workflows without proportionally increasing analyst effort.

Additional agents supported trend mapping, media monitoring, audience response analysis, and strategic research workflows across multiple client environments.

Evaluation and Observability

A major focus of the engagement was building the operational infrastructure required to safely run and improve agent systems over time.

Focused worked alongside Weber’s engineering teams to implement LangSmith for tracing, evaluation, and observability across the agent ecosystem.

Rather than relying on intuition-based QA or manual spot checking, the teams established structured evaluation workflows using trajectory testing, LLM-as-judge evaluation patterns, and trace-level inspection of agent behavior.

Focused also integrated evaluations directly into deployment workflows using LangGraph Platform and CI/CD infrastructure so agent quality could be validated continuously as prompts, models, and orchestration logic evolved.

This shifted evaluation from a post-deployment exercise into a core part of the engineering lifecycle.

Results

Focused delivered more than a dozen production agents that Weber’s strategists actively use in day-to-day workflows.

More importantly, the engagement helped establish a scalable engineering foundation for how Weber builds and operates agent systems internally.

Before the engagement, many workflows relied on brittle procedural logic and manually coordinated prompt chains that became increasingly difficult to maintain as complexity grew. Focused introduced orchestration patterns, evaluation infrastructure, and observability practices that gave Weber a more reliable framework for building long-running, tool-calling agent systems inside existing enterprise environments.

The engagement also helped establish a more collaborative operating model between strategists and AI systems.

Rather than replacing analysts, the platform accelerated research-heavy workflows while keeping humans embedded throughout the process for validation, refinement, judgment, and strategic interpretation.

Focused worked side-by-side with Weber’s teams throughout the engagement so the architecture, tooling, and operational patterns could continue evolving internally long after the initial delivery.