Inside the Spark Launch: Watching Productboard's Agentic Product System Run Live
For the past few years, every function in a company has gotten purpose-built AI. Sales, support, and engineering each have tools built specifically for how that work runs. Product management has had everyone else's leftovers: general-purpose chat tools that don't know your roadmap, your customers, or your codebase, and have to be re-briefed from scratch every time.
The cost shows up in how product managers actually spend their days. Not in the work itself (surfacing product opportunities, writing product specs, measuring impact), but in everything around it. Manually reading through feedback to find the signal. Copying a spec into a ticket and watching them drift apart. Tracking down what the team shipped last week to write Friday's status update. Work about work, and a lot of it.
On June 30th, Productboard's co-founder and CEO Hubert Palan and VP of Growth and Product Jordan Nolff ran the first public demo of Productboard Spark, live in their own workspace on real data, to show what product management looks like when that overhead disappears
Product Teams Finally Gets Their Own AI Moment
“It's one thing for a product manager to drop a call transcript into Claude and ask for the biggest pain points. It's another thing entirely to get every builder in your company pulling consistently high-quality signal out of hundreds of thousands of customer conversations.”
Hubert Palan
CEO & Co-founder, Productboard
Spark is Productboard's answer: what Hubert called the first agentic product system. At its core is an orchestrator that routes work across a team of specialist agents: a feedback analyst, a market researcher, a codebase analyst, a product expert, and a document author. Each is purpose-built for its slice of product work, grounded in what Hubert called a product data model: persistent memory of a company's customers, product strategy, and everything the team knows. Not a generic model that needs re-briefing every session.
That difference shows up in the work.
“I saved 1 week of work in just 90 minutes using Spark and successfully delivered the output to my executive team.”
Jay Smith
Product Owner, Vault
We Watched Spark Run the Entire Product Lifecycle
Jordan began the demo with a disclaimer that was also a bet: "Everything you see here is live in our own Productboard workspace. Same opportunities, same feedback, same specs we shipped last week. No sandbox, no fake data."
That choice is the whole reason to watch the recording. Seeing Spark think through a prompt in real time, on a workspace that wasn't staged for a keynote, is what makes it so compelling.Â
Here's the loop Jordan runs every day.
Spark Finds an Opportunities Buried in FeedbackÂ
Starting from Spark's home screen, Spark surfaces opportunities weekly from Productboard's full body of customer feedback. These are prioritized findings tied to company strategy, competitive context, and roadmap gaps. Every insight links back to the original customer note. Click a citation, and you're reading the source.
After selecting an opportunity, such as, “becoming the primary place product work happens”, Jordan can ask Spark by voice which underlying gaps the team wasn't addressing. While it runs, it references the agent knowledge layer: a standing set of context including strategy docs, personas, competitive intelligence, and brief templates that Spark references on every query, updated in real time across the whole workspace.
The response comes back with flagged gaps, each linked to live entities that he can click straight into, and five possible bets scored across three prioritization vectors. For example, a decision log, so teams stop losing the reasoning behind past calls when people leave or memory fades.
What used to take a half-day of reading feedback and cross-referencing the roadmap is now achievable in minutes.
The Spec Writes Itself Against the Actual Codebase
After identifying opportunities, Spark then drafts a product brief. It follows Productboard's standard template automatically, with every feedback citation embedded directly in the document. Hubert drew the contrast live:
“In Claude you can keep those references in a README file somewhere. Here it's attached to the entity, and the entities are related to each other because there's a rich structured data model underneath.”
Hubert Palan
CEO & Co-founder, Productboard
Once the product brief is complete, it’s time to generate a product spec. Spark pulls in a live analysis of Productboard's GitHub repositories, indexed across multiple repos and translated from code architecture into the language of user flows and product value, and builds the spec against what would actually work within the product’s existing constraints. The finished spec is attached directly to a new feature entity in the knowledge graph, feedback citations and all. No more spec that's already out of date by the time an engineer opens it.
From there, the team has a custom skill, trained on Hubert's actual Slack feedback history, that drafts inline comments the way Hubert would review a brief, making it easy to anticipate feedback before he reviews.
From Idea to Spec with Shipped Code in Under an Hour
Jumping into Claude Code, it’s easy to use Productboard's MCP server to pull the feature spec directly in, and build it in plan mode. During the demo, Jordan did this for a feature that went live in a staging environment in front of the audience.
From there, the team has a custom skill, trained on Hubert's actual Slack feedback history, that drafts inline comments the way Hubert would review a brief, making it easy to anticipate feedback before he reviews.
“This started from a cluster of customer feedback, and now I have working code and a prototype that I can feel myself. This entire flow took me under an hour.”
Jordan Nolff
VP of Product & Growth, Productboard
Spark ensures that product is not the blocker in a time when code can be delivered faster than ever. And it does so while incorporating all the context to invest in building valuable products.
A Same-Day Report Shows Whether the Feature Actually Worked
Once a feature has shipped, it’s time to measure the impact. Every morning, Spark generates a report pulling usage data from Amplitude alongside customer feedback from Productboard, producing an executive summary with an impact rating and its own confidence score.
That signal feeds back into the same feedback pipeline that generates new opportunities, bringing product makers back to the beginning of the product development lifecycle.
“The entire loop closed. Opportunity, spec, built, proving that it mattered and that we actually shipped value, not just slop. All in one place, all on our real data.”
Hubert Palan
CEO & Co-Founder, Productboard
Next Steps: Enable Spark Before July 31 and Register for Spark Forge
Now that you understand how Spark works, it’s time to try it with your own data! Every Productboard customer gets unlimited Spark usage through August 31st by enabling Spark by July 31st.
And to help everyone get building with Spark, Productboard is also running Spark Forge, a hands-on series built around the shift the demo was really about. There's a workshop for product leaders on July 8th and a four-part series for product makers starting July 9th. Be sure to save your seat.
We look forward to seeing the products you build with Spark!
Frequently Asked Questions About Productboard Spark
What is an agentic product system?Â
Software built specifically to run the full product development lifecycle: surfacing opportunities, writing specs, connecting to coding agents, and measuring impact, using specialized AI agents grounded in a company's real product data.
Is Spark safe to use with sensitive roadmap data?Â
Yes. Spark respects existing Productboard permissions and Team Space controls, runs under enterprise model agreements with zero day data retention, and doesn't train on customer data.
Does Spark replace tools like Claude, ChatGPT, or my coding agent?Â
No. It works alongside them. Spark connects directly to coding agents like Claude Code through MCP, and Productboard has been explicit that general-purpose tools still have a place in the workflow.
How much existing data does my workspace need before Spark is useful?Â
None to start. Spark pulls competitive and market context from the web on day one, and its specialized skills produce stronger output than a generic LLM prompt even before a workspace has accumulated feedback history.
