.png)
How multi-agent workflows save sales teams $125K/year
In the first episode of the Agno Builder Series, we spoke with Brandon Guerrero, Sr. GTM Engineer at The Kiln. He built a project called Playbook AI, an agent-driven workflow designed to take on the repetitive prep work that slows down sales teams.
In this post, we’re sharing a recap of that conversation for our reader-friends. We’ll start with the reality of how reps work today, then look at what Brandon built, how it runs on Agno, and what his final playbook looks like. If you’d rather watch the video, you can find it here.
The hidden workload behind every sales conversation
Anyone in sales knows just how much prep work is required before reaching out to a new account. Before an AE or SDR can confidently engage with a prospect, they need to:
- Learn what the company does
- Understand its audience, pricing, products, and competitors
- Identify which personas matter most
- Map their own product to the account’s needs
- Draft emails, talk tracks, and follow-ups
- Repeat this process for every account
Sales reps spend less than 30% of their time selling. The rest of their time is spent on admin and pre-outreach work. For a 10-person team, that adds up quickly. Brandon estimates it can reach around 2,500 hours a year. Playbook AI was created to take this burden off sales teams.
What the project does
Brandon describes Playbook AI in simple terms: it’s an intelligent sales playbook generator. It analyzes the vendor’s website and the prospect’s website. Then it uses that information to create the complete set of materials a rep needs to start selling.
The output includes:
- Personas and target roles
- Value props for each role
- Talk tracks and discovery questions
- Objections and grounded responses
- Email sequences with suggested timing
- Case-study-based proof points
Everything is tied directly to what the workflow extracted from the two companies’ sites. Reps get a playbook tailored to what that account cares about. They ask sharper questions and avoid generic discovery.
How the workflow runs behind the scenes
To understand the build, it helps to look at the steps the workflow follows. Brandon modeled it after the work an AE does, adding structure and automation.
1. Start with both domains
The workflow first checks whether the vendor and prospect websites are reachable. This acts as a simple check before the next step begins.
2. Scrape the homepages
Using Firecrawl, it pulls text directly from each homepage. This gives the workflow a high-level view of what each company does before digging into more content from each website.
3. Map the entire site
Next, it collects all URLs linked in each homepage, including pricing pages, case studies, features, security details, product pages, and more. This way, all relevant information gets collected, no matter where a company puts it.
4. Pick what’s most relevant
From that full map, the workflow identifies the pages most useful for understanding how the vendor’s product fits the prospect’s needs. Depending on the site, this could be dozens of URLs. Those pages are then batch-scraped for further analysis.
5. Extract structured information
Here, Brandon’s data models come into play, covering case studies, product offerings, value propositions, proof points, customer examples, and more. Specialist agents use those models to pull clean, structured information from the scraped text.
For example, if a company’s site mentions a case study, the workflow extracts the challenge, solution, and results into defined fields. The same happens for product details, personas, and objections.
6. Analyze the prospect
Another set of steps examines the prospect’s site, analyzing their messaging, pain points, target customers, and what they emphasize publicly. This information is brought into the workflow to help tailor the final playbook so it aligns with the prospect’s perspective.
7. Pull everything together into a playbook
Once the workflow understands both sides, it assembles the final output: persona-specific messaging, email sequences, objections and responses, talk tracks, and a battle card.
Brandon structured the workflow so that several steps run in parallel. There isn’t one giant agent doing everything. Instead, smaller agents focus on individual tasks. That keeps the process fast and avoids mixing contexts that don’t belong together.
Why Agno made this build possible
Brandon highlights two key features of Agno that made the process smoother. First, Agno makes it easy to create separate specialist agents, give them clear instructions, and run them in parallel.
Second, with Agent OS, deploying the workflow was simple. Brandon wrapped his logic in a serve.py file and had a FastAPI app running in minutes. He didn’t have to worry about infrastructure or exposing the workflow, which allowed him to focus entirely on the logic itself.
What the final playbook looks like
During the demo, Brandon shared an example targeting Salesforce. The workflow identified three personas: Chief Marketing Officer, VP of Sales, and Head of Customer Experience. It generated materials for each.
Each persona’s section included:
- What that role cares about
- Suggested talk tracks
- Discovery questions and follow-ups
- Common objections and grounded responses
- A structured, four-touch email sequence
The email sequences included objectives for each touch, proof points drawn from case studies, and guidance on when to send each message to improve response rates. All of this was generated from structured data extracted by the workflow.
How to build a similar agent using Agno
Brandon offers clear advice for builders: start with the quick start guide. You can also reference the examples in the Agno repo, but first, make sure you fully map out the manual version of your workflow. Automation comes after that.
See what’s possible for your sales team with Agno
Sales teams spend too much time collecting information and not enough time using it. Tools like Agno give builders a way to take the repetitive parts of this work and turn them into something reusable and consistent. To see exactly what Brandon built, watch the full video.