--- title: Software factory is the wrong bottleneck | daily.dev description: Everyone's building software factories. Meta's leaked numbers show that buys you 220% more code changes, 36% more features that reached users. For a small startup, shipping faster is important but the missing analyst, sales ops, product and marketer are even more important. Here's what our agent does in analytics, product operations, sales, marketing and engineering quality. canonical: https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/ og:type: article og:url: https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/ og:title: Software factory is the wrong bottleneck | daily.dev og:description: Everyone's building software factories. Meta's leaked numbers show that buys you 220% more code changes, 36% more features that reached users. For a small startup, shipping faster is important but the missing analyst, sales ops, product and marketer are even more important. Here's what our agent does in analytics, product operations, sales, marketing and engineering quality. og:image: https://media.daily.dev/image/upload/s--69JjuRYV--/f_auto,q_auto/v1/recruiter-landing/exec_f09854fa_ac1f_40dd_b489_d723feef6d0c_2bd7d6269a?_a=BAMAMiB80 og:site_name: daily.dev og:locale: en_US article:published_time: 2026-09-07 article:modified_time: 2026-09-07T14:03:53.130Z article:author: Ido Shamun twitter:card: summary_large_image twitter:site: @dailydotdev twitter:creator: @dailydotdev twitter:title: Software factory is the wrong bottleneck | daily.dev twitter:description: Everyone's building software factories. Meta's leaked numbers show that buys you 220% more code changes, 36% more features that reached users. For a small startup, shipping faster is important but the missing analyst, sales ops, product and marketer are even more important. Here's what our agent does in analytics, product operations, sales, marketing and engineering quality. twitter:image: https://media.daily.dev/image/upload/s--69JjuRYV--/f_auto,q_auto/v1/recruiter-landing/exec_f09854fa_ac1f_40dd_b489_d723feef6d0c_2bd7d6269a?_a=BAMAMiB80 --- Everyone is so busy building software factories lately and speeding up their development. StrongDM were the first to go viral over it, and now even big names like LaunchDarkly, Cloudflare, Vercel and Uber are publicly sharing their software factory journey. Now I get it, software factory is exciting. The very thought that every feature you have in mind can go live by burning some tokens is science fiction. And for those big corps it makes sense, because then they can shrink teams by up to 60% and hand the work to agents (yes, [Meta](https://thenextweb.com/news/meta-project-ot-ai-native-layoffs-cancelled), I'm looking at you). But for smaller teams that are low on budget and resources, is that the right way to consume your limited tokens? A software factory with no supporting roles is actually a slop factory, and not in the sense of the code it produces but the impact that will never come. Meta's own leaked numbers say it best: code changes went up 220% year on year, changes that reached users as features went up 36%. You can ship a hundred features a month now. Which one of them mattered? How do you even measure that? Where does the next token go, and how do you make sure the thing you shipped yesterday didn't quietly hurt production? Nobody is writing about that part, because it's not as fun as bragging about how many lines of code you shipped last week. An early stage startup usually consists of 2 founders, a few engineers and 1 sales/marketing/designer. These days, even without a software factory, engineers can move fast, very fast, but where they move and how they operate is the billion dollar question. For such teams the factory speeds up the one thing you were already fast at, and every other function is still one person or nobody. That's where agents change the math, and not only product wise, but in every domain. We're capped mostly by our imagination and our ability to articulate the constraints we face every day. A small team can think big and still win. Enough with the rant, let's get practical. Under the assumption of limited tokens and resources, here's how I think you should use your tokens as a small team, with what we actually do at daily.dev in each area. Quick context first. Most of what follows runs on Smith, our internal agent. It lives in our Slack and you talk to it like a teammate. It has access to pretty much everything: BigQuery, ClickHouse, Postgres, GrowthBook, our production API, GitHub, a browser and a scheduler for recurring jobs. Whatever it learns during a conversation, it writes back into a git repo we call the brain, so the next time someone asks for the same thing it already knows how. Today the brain holds 112 skills, all of them written by Smith itself, and only 7 of them are about writing or reviewing code. The rest is the stuff below. If you want the story of how we built it, I wrote about it in [We built an org-wide AI agent in 4 days. Here's what broke in the weeks after](https://daily.dev/blog/we-built-an-org-wide-ai-agent-in-4-days-heres-what-broke-in-the-weeks-after/?via=dailydev). You don't have to build your own, by the way. There are plenty of off-the-shelf agents now that plug into Slack, your warehouse and your repos. ## Analytics As a small team that is already post launch, you have analytics data coming in and A/B experiments running, and you wish to be more data driven than before. Unfortunately, you can't justify a full time data analyst yet, so your CTO does the high quality analysis and everything else is left unattended. This is fundamental for healthy decision making and your agent can help you with this. Give it access to your data warehouse and use it to run queries and bring you the right insights, so you can decide what's next. Remember to always ground its understanding in the code and how events are being triggered. In practice that means the agent reads the event definitions straight from the repos before it writes a single line of SQL, and rescans them when the code changes. Otherwise it will confidently count the wrong event and you'll make a confident wrong decision. Put cost guardrails on the warehouse too, because an agent that can run any query will eventually run a very expensive one. We use Smith to analyze user journeys, funnels, A/B experiments, release impact and much more. It's connected to BigQuery and GrowthBook as our primary sources, but it can also reach other secondary databases if needed. For experiments it does more than read the results. It creates the metrics with our SQL conventions, audits which feature flags are live with no experiment behind them, and writes the summary when an experiment is done so we can actually close it. In the last 90 days we concluded 50 experiments with a team of 14, and 6 are running right now. We recently ran a top of funnel analysis across our different channels (SEO, organic, referrals, AEO, etc) to understand where to double down as well as D1 retention analysis breakdown by application platform (extension, web, android, ios) and this is just the tip of the iceberg. That's the boring part of "data driven", but it's exactly what decides which of the hundred features was worth building. ## Product Operations You're at the point where user feedback is starting to pile up and you need to make sense of it. Some features require attention, moderation and maintenance. Trust and safety may already be a concern, and overall product operations is a thing, and you lack the person to handle it full time (at this point you probably see the pattern: missing functions in your team). Agents to the rescue! Personally, I think this is one of the areas most companies sleep on. It differs from one product to another, so I'll stick to our examples. We have a nightly cron job that finds spam users and acts on them, one of the biggest issues of any social platform. It cross-references posting patterns with account data, acts on the obvious cases through our internal API (the action is reversible, on purpose), and only escalates the edge cases to Slack for a human to look at. In the last 30 days that was about 1,700 accounts, roughly 80 a night. Pending posts for moderation go through the same shape of review, with per-channel rules for what counts as off topic. Our company verification process, which involves validating that the company exists, fetching metadata and a few more steps, is fully automated. It used to take someone a full day of review every 14 days, now it's done every day by an agent. Some weeks the queue is empty and the run finds nothing, which is fine. The point is that the job exists whether or not someone remembers it. Then there's content curation, like finding new sources for our feed based on supply and demand analysis, a weekly review of new tags we should add to the system, and more. It's like we have a small team of in-house editors and product people, where in fact we have none. ## Sales Everything here is highly correlated to money and has a direct impact on your revenue. At this point, the CEO is the only sales person (maybe there's one more), and they need to do sales along with the tons of other stuff early stage CEOs usually do. For a healthy sales and customer satisfaction pipeline, there's plenty of prep work to do. The agent can help you do extensive research to shortlist proper cold outreach targets and personalize it fully for each client. It can brief you before taking the call, monitor how the POC is going, send updates to the clients and check in on them. This is actually the biggest group of skills in our brain, bigger than engineering, and most of it belongs to one person. Every morning they get a brief for the day's meetings: who's on the call, what we know about them from our own data, the web and past email threads. At 6pm they get an end-of-day summary built from their inbox and the CRM. In between, the agent produces the campaign wrap reports, the renewal prep, the upsell emails with the real numbers from the campaign, the weekly status sheet across every active campaign and the monthly revenue forecast. For outbound, it sources and researches the accounts, drafts the sequences and classifies the replies into the CRM. Our rules are real research per account, small batches, and every step that spends money or sends an email goes through an approval gate. ## Marketing Marketing at an early stage startup is usually one person, or the CEO on Friday afternoons. Content and distribution are the parts agents are best at, and also the parts where they produce the most slop, so this is the section where I have to be careful. We run a few programmatic SEO sites next to the main product. Every headline passes a topic guard and a dedup pass so we don't publish the same thing twice under different titles, and a human sets the topics and the guardrails. Within those rules, our agent has access to Ahrefs and Search Console. It audits those sites for pages that cannibalize each other's keywords and opens the PR with the redirects, which is mostly cleaning up our own past slop. It keeps the headline backlog topped up, with a topic guard so our recruiter content doesn't leak onto the ads site. It writes and audits the titles and descriptions of our tag pages, and it builds our disavow file from our own spam detection data, which is a nice example of one automated role feeding another. Then there's AEO, which is SEO for LLMs. We track where daily.dev gets mentioned or cited in ChatGPT, Claude and the rest, measure how much of our traffic actually comes from AI agents, and look for the content gaps. And it handles the boring recurring stuff: the changelog aggregated from both repos, the release notes, and the first draft of the monthly product email. Most of it would never have happened without an agent. ## Engineering quality Keeping your product up, running and in its best shape requires attention. Have your agents monitor production, escalating or even solving production issues, or potential ones. We have cron jobs to detect missing indexes, slow queries and even cost optimization suggestions. The same thinking applies before anything reaches production. We have a PR reviewer per repo, grounded in three years of our own human review comments. When a risky change ships, we schedule a measurement job that compares before and after against a control and posts the result. ## Who watches the agents One warning before the conclusion. It's not as perfect as it may sound. Agents are still likely to generate slop, and you have to nurture and groom them constantly. Automate gradually until you build confidence in their output. Give them feedback, adjust, and help them learn where they make a mess versus where they provide actual value. ## Conclusion So where should you spend your tokens? You probably know better what blocks your business from growing, but it's not necessarily shipping faster. Shipping is the easy part now. Deciding where to ship, and keeping everything around it running, is where your tokens actually compound. It's time to think big and make these dreams come true with agents. On this happy note, back to building! ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/og-image.png?v=a830cdf1","width":1200,"height":630},"sameAs":["https://twitter.com/dailydotdev","https://www.linkedin.com/company/dailydotdev","https://github.com/dailydotdev","https://www.instagram.com/dailydotdev"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","description":"Free, personalized developer news aggregator. Stay on top of software development news, AI coding tools, and web dev - curated daily from trusted sources.","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}},{"@type":"WebPage","@id":"https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/","url":"https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/","name":"Software factory is the wrong bottleneck | daily.dev","description":"Everyone's building software factories. Meta's leaked numbers show that buys you 220% more code changes, 36% more features that reached users. For a small startup, shipping faster is important but the missing analyst, sales ops, product and marketer are even more important. Here's what our agent does in analytics, product operations, sales, marketing and engineering quality.","inLanguage":"en-US","isPartOf":{"@id":"https://daily.dev/#website"}},{"@type":"Article","@id":"https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/#article","headline":"Software factory is the wrong bottleneck","url":"https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/","datePublished":"2026-09-07","dateModified":"2026-09-07T14:03:53.130Z","isPartOf":{"@id":"https://daily.dev/#website"},"publisher":{"@id":"https://daily.dev/#organization"},"mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/"},"description":"Everyone's building software factories. Meta's leaked numbers show that buys you 220% more code changes, 36% more features that reached users. For a small startup, shipping faster is important but the missing analyst, sales ops, product and marketer are even more important. Here's what our agent does in analytics, product operations, sales, marketing and engineering quality.","image":{"@type":"ImageObject","url":"https://media.daily.dev/image/upload/s--69JjuRYV--/f_auto,q_auto/v1/recruiter-landing/exec_f09854fa_ac1f_40dd_b489_d723feef6d0c_2bd7d6269a?_a=BAMAMiB80"},"author":{"@type":"Person","name":" Ido Shamun","url":"https://app.daily.dev/idoshamun"},"potentialAction":{"@type":"ReadAction","target":"https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/"}},{"@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://daily.dev/blog/"},{"@type":"ListItem","position":3,"name":"AI","item":"https://daily.dev/categories/ai/"},{"@type":"ListItem","position":4,"name":"Software factory is the wrong bottleneck","item":"https://daily.dev/blog/software-factory-is-the-wrong-bottleneck/"}]}]} ```