Agentic AI & Vibe coding

Background

A fast-growing regional marketing agency wanted to move beyond AI as a buzzword and embed it into the daily workflows of its campaign, strategy, and operations teams. The agency faced mounting complexity in managing multi-channel campaigns, producing client-ready reports, and coordinating internal operations, all while trying to protect margins in the face of rising SaaS and data platform costs. Individual staff were experimenting with AI assistants for ideation or copy tweaks, but there was no shared approach for turning these experiments into robust internal tools and automations. Humanityze was engaged to design and run a guided AI buildathon and agentic automation program tailored specifically to marketing workflows, using platforms such as Replit, Claude Cowork, OpenAI Codex, and no-code integration tools.

Defining the problem

The agency’s teams were spending significant time on repetitive, manual tasks across campaign monitoring, reporting, and coordination, which crowded out higher-value strategic and creative work. Off‑the‑shelf tools for media monitoring, email campaigns, and analytics were both expensive and inflexible, and teams often reverted to spreadsheets and manual copy‑paste processes to bridge gaps between systems. Although staff had some exposure to AI assistants, usage was fragmented: a few “power users” were pushing the boundaries, while many others were uncertain about where to start or concerned about quality, brand safety, and client perception. There was no systematic way to identify high‑impact AI opportunities, prioritize them, and then design automations that could be monitored, governed, and scaled across accounts and regions. Leadership needed a way to create visible quick wins, build internal confidence, and ensure that any AI-powered tools were grounded in real marketing pain points rather than technology for its own sake

Purpose of Humanityze’s involvement 

Humanityze’s role was to convert scattered interest in AI into a structured capability-building journey that produced working tools, clear ownership, and a repeatable internal playbook. The engagement was built around two linked half‑day sessions: a “vibe coding” buildathon to introduce AI-assisted building and rapid prototyping and an “agentic AI” workshop focused on autonomous agents and multi-step workflow automation. Before the sessions, Humanityze ran a technical assessment survey to understand participants’ familiarity with tools like ChatGPT, Claude, Replit, and no-code platforms, then used this data to group people by experience level and role. Working with agency leaders, Humanityze curated a set of marketing-centric opportunity areas—such as custom media and social monitoring, campaign automation, client reporting, and onboarding workflows—to ensure teams were solving problems with clear business value.

The outcome 

In the first session, cross-functional teams learned to use conversational AI together with OpenAI Codex and low-code tools to design and build simple but usable prototypes. Using Replit as a lightweight development environment, teams created custom monitoring scripts and scrapers that pulled data from media and social sources, while Codex assisted with generating and refactoring code, handling boilerplate, and translating natural language requirements into working functions. Participants also built campaign-support tools such as automated email and asset generators, as well as report-building helpers that aggregated multi-source data into template-based outputs ready for polishing. Each prototype was documented with “how to use this” guides and demonstrated in short business-focused demos, making it easy for non-technical leaders to see direct relevance and potential ROI.

The second session shifted the focus from tools to end-to-end workflows, introducing agentic concepts using platforms such as Claude Cowork and Replit-based agents. Participants mapped their personal and team workflows—daily digests, client update routines, inbox triage, campaign status checks—and then designed agents with triggers, guardrails, and escalation rules, integrating them with email, Docs, Sheets, Slack, and other collaboration tools. Using Claude Cowork, teams designed agents capable of summarizing inbound information, drafting updates, and deciding when to hand work back to humans based on predefined confidence and risk thresholds, while Replit hosted the orchestration logic and external API calls. Across the two sessions, the agency produced six to ten working tools and agents that addressed real operational pain points, with early modelling indicating that reduced vendor dependence, faster reporting cycles, and reclaimed staff hours could more than pay back the program within a few months. Just as importantly, a core group of participants emerged as internal AI champions, comfortable with platforms like Replit and Codex and able to collaborate with non-technical colleagues using shared workflow maps and playbooks. 

Next steps

Post‑workshop, the agency and Humanityze agreed on a phased follow-through plan to ensure the prototypes and agents moved into production use rather than remaining one-off experiments. Governance guidelines were drafted to clarify where Claude Cowork and other AI agents could act autonomously, when human review was mandatory, how data should be handled, and how exceptions and edge cases would be monitored. A summary report and prototype inventory were shared with the broader organization, along with simple usage guides and success stories to accelerate adoption and inspire additional ideas.