Skip to content
Software Developments, Dev Trends & Reviews

News

Seven people, 2,500 articles a month: Featured keeps its hard guarantees in Postgres, not its workflows

The Inngest Featured case study shows how seven people ship 2,500 articles a month with cron sweeps, durable steps and a Postgres uniqueness constraint.

SNSoftware News5 min readBased on a press release from Featured

news/inngest-featured-case-study-orchestration.md1,105 words

The Inngest Featured case study describes how a seven-person team runs HARO, Connectively and Featured, shipping 2,500 articles a month without anyone operating the pipeline by hand. Its lesson for small engineering teams is where each guarantee lives: schedules in the orchestrator, hard rules in the database.

Key points

  • Featured says seven people now run its three brands, work that in 2020 took eight people coordinating it by hand.
  • The pipeline ships 2,500 articles a month, according to the company.
  • A cron job checks every 15 minutes for questions whose current deadline has passed, instead of a timer sleeping until a fixed deadline.
  • Each closed question gets one run of 60 to 150 durable steps, and Inngest handles all 45 of the company's cron jobs.
  • A Postgres uniqueness constraint, not workflow code, stops an expert from being notified twice about the same article.

It says Featured replaced a human coordination team with software that schedules, sequences and stops its article pipeline. Inngest, which sells durable execution for background jobs and AI workflows, published the customer story on its site, and Featured's newsroom announced it on November 18, 2026.

Connectively, one of Featured's three brands, lets publishers post a question to fill a gap on their site. Experts answer it before a deadline, and the system then picks answers, assembles an article and pushes it to the publisher's site. Inngest's account puts the average at 42 answers per question.

Software News is published by Connectively, so this article covers our own parent company.

Why use a cron sweep instead of a durable timer?

Because a sweep reads the deadline as it stands now, while a timer remembers the deadline it was created with. Featured says publishers move deadlines back, and the sweep needs no special handling for that.

According to the case study, the trade-off is deliberate: a run can start up to 15 minutes after a deadline, but it can never close a question early and throw away answers still arriving. Inngest supports both patterns; Featured chose the sweep.

Once a question closes, exactly one run starts. Evaluation reads each expert for experience, expertise, authority and trust, then admits answers one at a time, checking each against those already chosen so the article does not repeat itself. Inngest's write-up says that loop runs up to 25 times.

Orchestration decides what happens and when. The database decides what's allowed to happen at all.

— Brett Farmiloe, founder and CEO, Featured, as quoted by Inngest

Two things: the rule that an expert hears about an article only once, and the decision to keep trying after a bad result.

The first is a uniqueness constraint in Postgres, which refuses a second notification row for the same expert and article however the workflow behaves. That puts the guarantee in a system most teams already run. In the 2025 Stack Overflow Developer Survey, 55.6% of respondents said they had done extensive work in PostgreSQL over the past year, more than any other database, as did 58.2% of professional developers. For most teams, the same pattern needs no new infrastructure.

The second is that most functions do not retry. Featured says a run that cannot produce an article it would stand behind marks itself failed and stops, because publishing a weak article under a publisher's name is worse than a late one.

Our analysis: what small teams can take from it

The pattern is not specific to Inngest. Teams building agent and content pipelines can split the work three ways: a database for state and the rules that must never break, a scheduler for timing and flow control, and a compute platform for the model calls. Featured describes Postgres on Supabase, Inngest and Vercel in those roles.

The cron-sweep choice suits any job whose trigger date can move, such as renewals, deadlines or approvals: the job reads the current date each time instead of holding a timer that has to be cancelled and reset. Turning off retries suits any step whose output reaches a customer directly, as long as each run records its status, so a failure shows which stage stopped and nothing after it ran. According to Inngest, Featured's team watches an internal dashboard built on those job records.

The staffing point matters as much as the code. In the case study, Featured credits the orchestration layer with moving its people from coordination to judgment work, such as shaping questions and tuning how answers are evaluated. For more on how teams coordinate agents, see our reports on an orchestrator for AI agents, agents taking on more tasks in Attio and the tools teams use to improve workflow.

Frequently asked questions

Which stack does Featured run its article pipeline on?

Featured says its state lives in Postgres, using Drizzle on Supabase, while Inngest handles scheduling, events and rate limits, and Vercel runs the pipeline and its model calls.

How long can a Connectively run wait after a deadline?

Up to 15 minutes, since the sweep runs on that interval. It never starts before the deadline.

How big are HARO and Connectively?

Featured says HARO sends more than 100 million emails a year to over 125,000 sources, Connectively has passed 100,000 expert members, and the two have fielded more than 2.5 million pitches.

What counts as one article in Featured's pipeline?

One publisher's question, closed at its deadline, with the expert answers the system selected assembled into a post and pushed to that publisher's site.

Sources