Introducing: Gillnet

There is a temptation when building an AI product to start with the final architecture.

Design the SaaS. Build the multi-tenant backend. Add the dashboard. Add OAuth. Add an agent. Add queues, permissions and infrastructure. Then hope the underlying workflow is actually useful.

With Gillnet, we did the opposite.

Starting Q2 2026, DitchNow worked with Jonathan Garcia and Solvism to build and productize Gillnet, an AI-assisted sourcing product for recruitment.

The project moved through three distinct architectural stages:

→ Gillnet Lite: n8n workflows + Slack-only UI
→ Gillnet v1.0: real backend + Slack workspace app
→ Gillnet v2.0: multi-tenant product + Slack OAuth + Web UI + embedded AI assistant

It shows why the job is not simply to "build the AI feature." The harder work is deciding what to validate first, what architecture is appropriate at each stage, and when a prototype has earned the right to become software.

The operational problem came first

Solvism had something more valuable than a feature list: recruitment domain expertise and a clear operating workflow.

The sourcing process needed to move from a vacancy to a set of candidates, evaluate those candidates, let a recruiter review the decisions, prepare outreach and keep a human in control before messages were sent.

In simplified form:

Vacancy → Search → Enrichment → Scoring → Candidate Review → InMail Generation → Approval → Outreach

The technical challenge was to make that loop work reliably without prematurely turning it into a large SaaS build.

So the first objective was not "build Gillnet v2.0."

It was: prove the workflow.

Phase 1: Gillnet Lite - proving demand with n8n and Slack

We built the original Gillnet implementation from the ground up as a set of n8n workflows.

Slack was the only user interface.

That was deliberate.

At this point, we needed fast iteration around the sourcing logic, provider integrations, AI-assisted candidate evaluation and the human review steps. Building a large application shell before those workflows were understood would have created unnecessary product and engineering cost.

The Slack-first version let us test the real operational sequence:

• vacancy intake and parsing;
• candidate search and enrichment;
• AI-assisted scoring;
• recruiter review;
• InMail generation;
• manual approval;
• outreach execution.

It also exposed where the real complexity lived.

The important lesson was that the workflow itself had become valuable enough to deserve a stronger application architecture.

That is when the prototype architecture had done its job.

Phase 2: Gillnet v1.0 - move beyond the workflow engine

One of the easiest engineering mistakes is to keep extending a prototype because it already works.

At some point, the speed gained from a workflow tool becomes the constraint.

For Gillnet, that point meant moving away from an n8n-first system and rebuilding the product around a real application backend.

Gillnet v1.0 introduced the things the workflow prototype should not have been responsible for long term:

• persistent campaign and candidate state;
• background workflow execution;
• clearer campaign lifecycle management;
• provider integrations behind application boundaries;
• proper recruiter review state;
• a Slack workspace application as a product surface rather than an automation endpoint.

Slack remained central to the user experience, but underneath it the system had become an application.

That architectural shift mattered more than adding another AI prompt.

Phase 3: Gillnet v2.0 - productized for broader use

The next question was different:

How do you take a working internal application and make it capable of operating as a broader product?

Gillnet v2.0 added the infrastructure and product surfaces required for that transition.

Multi-tenancy

The application had to stop assuming that one workspace represented the entire world.

Tenant-aware data and application boundaries were introduced so Gillnet could support multiple organizational contexts rather than a single internal deployment.

Slack OAuth and multi-workspace installation

A Slack application is easy to demo in one workspace.

It is a different product problem to make that application installable and usable across workspaces.

Gillnet v2.0 added Slack OAuth and the workspace-aware installation flow required for broader distribution.

Redesigned Slack experience

The Slack surface evolved into an operational product interface: campaign state, review queues, candidate decisions, message generation and workflow actions became part of a clearer recruiter experience.

Web UI

Slack is useful, but it should not be forced to carry every workflow.

Gillnet v2.0 added a Web UI for operational visibility, review queues and pipeline-level interaction. Slack and the Web application became two surfaces connected to the same underlying workflow state rather than separate implementations.

Embedded AI assistant

AI assistance was brought into the application itself rather than remaining an invisible background process. The assistant could participate in the product workflow while the system retained explicit human decision points.

The production complexity was outside the prompt

The most interesting work in an AI product is often everything surrounding the model.

For Gillnet, that included:

• deciding when the n8n architecture had reached its limit;
• preserving validated sourcing logic while replacing the orchestration layer;
• introducing persistent application state;
• designing tenant boundaries;
• handling asynchronous sourcing and review workflows;
• making Slack authentication and installation workspace-aware;
• connecting Slack and Web to one product state;
• keeping human approval in front of outreach;
• packaging the system for production-oriented deployment and handover.

Those are product engineering decisions, not prompt engineering decisions.

What Solvism owned, and what DitchNow owned

Jonathan Garcia and Solvism brought the recruitment expertise, sourcing knowledge and operating context.

DitchNow translated that domain knowledge into product architecture and working software.

Vacancy pipeline is built based on Jonathan's years of experience talking to 850+ candidates and filling 160+ vacancies. All of the vacancy comprehension, utilization, parsing, search and creating readable InMails were built based on his taste and communication style.

Across the engagement, DitchNow owned the technical evolution of the system: workflow architecture, backend direction, Slack application architecture, Web UI delivery, multi-tenancy, Slack OAuth, AI assistant integration and the path through production.

Our role was not to replace the domain expert. It was to give the domain expert a technical ownership layer capable of turning that expertise into a product.

That's why, Gillnet achieved 95% accuracy in candidate search in every internal and external tests.

From internal workflow to market-facing product

Gillnet is already being used in a customer context. It is currently in closed beta and will be released officially in the SOSU EU on Oct 7.

Jonathan has publicly introduced Gillnet as Solvism's sourcing product and announced that Gillnet is the primary sponsor of Sourcing Summit Europe in Amsterdam on 7–8 October 2026.

You can read more here 👉 https://www.linkedin.com/posts/jonathanmcgarcia_gillnet-sosueu-sourcing-activity-7505266421354962944-bLuY

The bigger lesson: product should follow evidence

Gillnet is a good example of how DitchNow approaches AI product delivery.

We do not need a client to arrive with a perfect technical specification or a final architecture diagram.

The path we follow is:

1. understand the operational problem;
2. build the smallest system that proves the workflow;
3. learn where the real complexity sits;
4. preserve what works;
5. replace what no longer scales;
6. evolve the system into production software.

The ability to move between those stages is more valuable than picking a sophisticated stack on day one.

If you are building an AI-heavy product and need someone to own the system around the intelligence - workflows, integrations, application architecture, product surfaces and production delivery - talk to DitchNow.

Case study: Gillnet Casestudy
DitchNow: https://ditchnow.com