How Omnia Agent Took a New Product to #1 in AI Search

I'm a software engineer at Omnia and I know almost nothing about real-life marketing. In July I started using the thing I'm building, Omnia Agent, on a friend's new open-source project, PrintStash (a 3D print file manager), whose website had just launched, and I spent 5 to 10 minutes most days deciding what to approve from its proposals. By late October, PrintStash was the brand AI engines mentioned most when people asked about managing 3D printing files, ahead of established projects with years of history and large communities, such as OctoPrint, Mainsail, Fluidd and Manyfold.

Try it yourself

• Googlebest 3d print file manager (ask Google, opens in a new tab)
• ChatGPTWhat is the best self-hosted 3D print file manager? (ask ChatGPT, opens in a new tab)

Answers vary by person, place and day.

Where PrintStash started

PrintStash is my friend's project: a self-hosted web app that helps with 3D printing workflows and file management.

The first thing to decide was which prompts to monitor, and I froze in front of a blank page. I had never chosen prompts for a brand, so instead of writing them myself, I worked them out in a conversation with Omnia Agent. I put myself in the shoes of someone printing at home and described the conversations they would have with ChatGPT about their 3D print files, such as how do I organize thousands of STL files?, how do I send G-code to several Klipper printers? or what's the best self-hosted file manager for 3D printing?, and the agent turned them into AEO prompts that follow Omnia's best practices.

Most of those people have never heard of PrintStash, and many don't know a tool like it exists. So I bet on showing up in those conversations, while people are still describing their problem and before they know which tool to look for. In AEO terms, these are non-branded prompts, because they don't name PrintStash, and most of them sit early in the funnel, where people explore a problem rather than compare products they already know.

When monitoring started on July 8, AI engines answered them with OctoPrint, Mainsail, Fluidd and Manyfold, long-established projects with big communities, and mentioned PrintStash in only 7.7% of their answers in the first two weeks.

In its first days on the project, Omnia Agent proposed 14 new posts targeting those prompts, which went live on the site, and one of them, a roundup of the best self-hosted 3D print file managers, became the most cited page on printstash.org, with 3,454 citations in AI answers from September 26 to October 26.

PrintStash went from barely mentioned to #1 in AI visibility

Hover, or focus the chart and use the arrow keys, to read any day.

• PrintStash
• OctoPrint
• Mainsail
• Fluidd
• Manyfold

See the data

Date	PrintStash	OctoPrint	Mainsail	Fluidd	Manyfold
Jul 8	1.7%	30.3%	27.7%	24.4%	37.0%
Jul 15	6.0%	39.9%	29.4%	27.6%	30.1%
Jul 22	10.2%	41.1%	32.2%	28.9%	28.9%
Jul 29	11.2%	41.4%	32.0%	29.7%	29.0%
Aug 5	14.6%	40.0%	30.9%	28.9%	27.5%
Aug 12	21.5%	38.6%	32.6%	30.8%	27.3%
Aug 19	23.9%	41.3%	34.0%	31.5%	25.6%
Aug 26	26.3%	44.0%	36.4%	33.4%	21.0%
Sep 2	32.0%	43.5%	38.6%	35.9%	20.5%
Sep 9	35.6%	44.1%	36.2%	34.4%	22.7%
Sep 16	42.1%	39.1%	33.2%	30.6%	23.3%
Sep 23	45.7%	39.7%	33.6%	30.9%	22.3%
Sep 30	49.7%	38.7%	32.7%	30.5%	22.0%
Oct 7	48.1%	39.4%	33.6%	31.5%	21.9%
Oct 14	48.9%	38.4%	34.6%	32.4%	22.8%
Oct 21	50.5%	37.7%	34.2%	31.4%	24.1%
Oct 26	50.3%	37.8%	33.2%	30.3%	24.0%

Most of the climb happened between mid-August and late September. PrintStash passed OctoPrint, its #1 competitor, on September 14 and stayed first through October.

My part was the direction, not the grunt work

Omnia Agent researched, proposed, wrote, shipped and measured, while I set the direction and decided what was worth publishing under PrintStash's name. The agent carried out every one of the 137 Projects completed by October 26 from start to finish.

The agent's scheduled tasks ran during the night, so when I started working in the morning everything was waiting for my input: new proposals, measurement results and a weekly report each Monday. I discarded the proposals that didn't fit, adjusted and approved the ones that did, and went back to my engineering job while the agent wrote the pages, checked every claim against the latest product release and published them.

It wasn't always ten minutes a day; in late August and September I spent several long sessions building the workflow described below, mostly because building automation is the part I usually enjoy the most, but also because I knew the efforts would pay off.

Most of my time went into taste, in the form of a few rules the agent followed:

• Use a developer-centric voice, without em dashes, inflated vocabulary or typical LLM writing patterns.
• Treat the product itself as the source of truth, and check every claim on the site against the latest release.
• Ship every language in sync, so a change to one page goes out in all supported languages at once.
• Be ambitious with content, because writing is cheap for us, but never pad a batch to hit a number.
• Don't cold-pitch blogs, which I stopped in October after it produced almost nothing.

The rule that made me trust it

The second rule, treating the product as the source of truth, is what made me trust the whole workflow. The agent takes its facts from the product's code, so every claim on the site, from what a feature does to what PrintStash can't do yet, is backed by how the product actually behaves instead of by what a language model fabricates. Without that, I wouldn't have let AI-written pages ship under PrintStash's name.

The rails Omnia Agent runs on

I'm an engineer, so I treated this as a system to build, with Omnia Agent as the main worker.

1. Jul to mid-AugCopilotI asked the agent what to do next and it answered with a plan, which I often pasted into Claude Code to write and publish the page, so content was live about five minutes later. Later, as Omnia Agent got better at writing content, it took over that part too.
2. Mid-Aug to late SepMy custom pipelineAgent skills (written procedures the agent loads) and scheduled tasks ran a work board I kept in Notion: proposals three times a week, automated execution of anything I marked as approved, measurement at 7, 14 and 30 days, and weekly audits of the blog and the site.
3. Since Sep 29Omnia ProjectsOmnia shipped native Projects and I moved everything my Notion board held into them. Projects don't cover measurement yet, so I kept a small custom measurement workflow on top of the agent for that part. With the tracking handled by the product, I spent less time in the platform while shipping more work.

The tooling went through these three phases, but the underlying shape of the work didn't change.

The AEO loop my work followed

Back to Observe, with what the last round taught us

1. 01Observe
2. 02Diagnose
3. 03Recommend
4. 04Act
5. 05Measure

Back to Observe, with what the last round taught us

1. 0101ObserveOmnia monitors 154 non-branded and 52 branded prompts every day across six AI engines, in English for the US and Spanish for Spain.
2. 0202DiagnoseThe agent reads the aggregated answers to see who gets named, which pages get cited and what engines say PrintStash lacks.
3. 0303RecommendOmnia's Project Suggestions, plus my own scheduled task when I want more, propose evidence-backed Projects, and scheduled weekly blog and site audits propose fixes as new Projects.
4. 0404ActThe agent writes new pages or edits existing ones in every language the site supports, checks every claim against the latest product release, publishes them and checks the live pages.
5. 0505MeasureThe agent measures each change 7, 14 and 30 days after it ships, comparing its target prompts with how they performed before, and feeds the outcome into the next round of the loop. Each change is treated as a hypothesis, since there are too many variables to be sure it's what moved the numbers, but the result is a signal worth having.

Back to Observe, with what the last round taught us

The scheduled tasks

Omnia Agent runs these automatically, so the results are ready before I start the day.

Most of them are thin wrappers over a skill that is already built into Omnia Agent, listed in the second column (a dash means no built-in skill covers the task yet). For Generate Projects, the built-in Omnia capability is Project Suggestions rather than a skill. The built-in skills are interactive: before they start, they ask you what to cover and where to focus. My wrappers answer those questions in advance, so the same work can run unattended, with nobody watching.

Task	Built-in skill	When	What it does	Why
Generate Projects	—	Mon, Wed, Fri 08:30	Researches current gaps in AI answers and creates up to three evidence-backed Projects.	Gives me more Projects to work on when Omnia Project Suggestions aren't enough.
Weekly brand report	create-aeo-report	Mon 09:00	Emails me a PDF with what moved, on which prompts and engines, and which pages were cited.	A short executive summary, so I never spend more than 10 minutes a day on this.
Blog curator	—	Tue 09:00	Checks every claim on the blog against the latest release and fixes the facts that are out of date.	Stops engines from repeating outdated facts that could undermine PrintStash or the site's standing as a source.
Content decay	rescue-decaying-content	Wed 09:00	Finds pages that are losing citations and opens refresh work for the ones worth saving.	Wins back citations before a page drops out of the answers.
Site audit	site-audit	Thu 09:00	Checks crawlability, structured data, language versions and page speed, and escalates regressions.	Catches SEO, crawlability and performance regressions before they cost citations.
Cited URL health	—	Thu 10:00	Checks every page the engines cite and flags broken pages and pages that block AI crawlers.	Keeps the most cited pages from turning into broken links and losing their citations.
Prompt audit	audit-prompt-keywords	1st of the month	Reviews the monitored prompts, proposes which to retire and which demand is missing.	So I can maximize the impact the monitored prompts' data provides and minimize monitoring costs.

How a change reaches the site

Approving a Project isn't pressing a button and letting the magic happen: my approval decides what gets built and how far it goes, and from there every change goes through the same steps, so one that fails the repository's quality checks doesn't ship until it's fixed. The steps are mechanical, but the judgment behind them is human, because they apply the rules above, written down once and followed every time.

1. Re-check the premise against today's answers and the live site, so the work isn't outdated or already done somewhere else on the site.
2. Check every product claim in the draft against the latest product release.
3. Write the English page, then localize it to every other language the site supports.
4. Run the site's quality checks, so there are no technical or performance regressions on the site or on the changed pages.
5. Publish, and open the live page to validate the changes.
6. Resubmit the sitemap and notify search engines.
7. Record the ship date as the input for the measurement step.

What the data says

Across all the monitored non-branded prompts, PrintStash ranks #1 on four of the six engines I track, and Google AI Mode and Gemini are the two where it isn't, at 4th and 5th.

#1 on four out of six engines

• Jul 8 to 21
• Sep 26 to Oct 26

See the data

Engine	Jul 8 to 21	Rank	Sep 26 to Oct 26	Rank
Copilot	11.6%	6th	80.1%	1st
Perplexity	2.6%	23rd	69.8%	1st
Google AI Overviews	1.4%	52nd	46.5%	1st
ChatGPT	9.1%	8th	46.3%	1st
Google AI Mode	0.8%	81st	33.9%	4th
Gemini	3.5%	37th	22.6%	5th

The biggest gains came on Copilot and Perplexity, while Gemini and Google AI Mode moved least, a pattern that shows up again in what worked and what didn't.

Citations tell the same story from the other side: by October, the website engines cited most when answering these questions was printstash.org, ahead of GitHub, Reddit and YouTube, which dominated in July.

printstash.org became the most cited site

Hover, or focus the chart and use the arrow keys, to read any day.

• printstash.org
• github.com
• reddit.com
• youtube.com

See the data

Date	printstash.org	github.com	reddit.com	youtube.com
Jul 8	0.0%	8.8%	9.0%	6.7%
Jul 15	0.3%	10.0%	8.5%	6.9%
Jul 22	1.2%	9.9%	11.0%	7.2%
Jul 29	2.6%	11.2%	10.8%	7.0%
Aug 5	3.3%	9.8%	13.6%	8.1%
Aug 12	4.6%	9.4%	11.3%	6.9%
Aug 19	5.3%	11.7%	9.3%	6.8%
Aug 26	8.3%	13.3%	8.9%	6.3%
Sep 2	14.2%	11.9%	8.0%	5.4%
Sep 9	16.8%	11.2%	8.4%	5.4%
Sep 16	19.2%	10.6%	7.5%	4.1%
Sep 23	21.4%	9.9%	7.6%	4.7%
Sep 30	22.2%	10.1%	7.1%	4.5%
Oct 7	26.2%	9.2%	7.0%	4.6%
Oct 14	22.8%	13.2%	4.9%	3.2%
Oct 21	22.3%	13.7%	3.9%	3.1%
Oct 26	21.8%	13.5%	3.7%	3.0%

printstash.org passed github.com on August 31 and led from then on, while Reddit's share fell by more than half.

The question I care about most

Every prompt I monitor matters, but to succeed I realized I had to set clear priorities for everything I did. So I picked the prompt that puts the brand's name closest to the conversion, What is the best self-hosted 3D print file manager?, and everything I worked on was meant to push PrintStash higher and higher on it.

From September 26 to October 26, AI engines mentioned PrintStash in 95.5% of their answers to What is the best self-hosted 3D print file manager?, and printstash.org was the most cited site for that question, with 32.3% of all citations.

Google's AI Overview names PrintStash first

On this one question, PrintStash ranks first or second on every engine, and printstash.org is the most cited site on every engine except ChatGPT, where GitHub wins and printstash.org sits second.

PrintStash and printstash.org, engine by engine

Engine	Mentioned in	Rank	printstash.org citation share
ChatGPT	100%	1st	24.8% (2nd)
Copilot	100%	1st, tied with Manyfold	51.9% (1st)
Google AI Overviews	96.7%	2nd, after Manyfold	27.2% (1st)
Google AI Mode	96.7%	2nd, after Manyfold	29.6% (1st)
Perplexity	90.0%	2nd, after Manyfold	32.4% (1st)
Gemini	90.0%	2nd, after Manyfold	37.4% (1st)

What measuring every change showed

Every change had a baseline when it was proposed and measurements 7, 14 and 30 days after it shipped. Those measurements covered visibility and share of voice on the change's target prompts, citations of the exact page, and whether the answers named PrintStash or only cited printstash.org without mentioning the brand.

Based on them, the agent classified the change into one of four outcomes:

• Improved: its target prompts had a meaningful improvement, and the change is the likely reason, but it is not certain.
• No change: its target prompts didn't show a meaningful improvement.
• Unclear: its target prompts had a meaningful improvement, but something else could explain it, such as another page shipped at the same time or an engine update.
• Worsened: its target prompts got meaningfully worse.

69% of 141 measured changes improved

• 69% improved (97)
• 18% no change (25)
• 13% unclear (19)
• 0% worsened (0)

How often each kind of change improved

• New pages that answer one specific question51 of 65 78%
• Fixing wrong sentences on PrintStash's own site11 of 14 79%
• Rewrites and refreshes of existing pages28 of 45 62%
• Technical and structured-data work2 of 7 29%
• Standalone release posts0 of 3 0%

New pages that answer one specific question and fixing wrong sentences on PrintStash's own site drove most of the improvements. Results also took time to show: of the 71 changes shipped in August that reached their 30-day measurement, 76% improved, 11 of which read "no change" at 7 days but "improved" at 30.

What really worked

• One page per real question, with the answer and the brand in the first sentence. Each page targets one question buyers ask and answers it in its first line, with PrintStash as the subject.

  For example, for the prompt How do I send G-code to my Klipper printer straight from my file library?, the agent wrote Send G-code to a Klipper printer from your library, which opens with "PrintStash sends a G-code revision from your library straight to one Klipper printer". Pages like this got their first citation 4 to 10 days after shipping and most of their citations in weeks 3 and 4.

• Rewriting the opening of pages engines already cite. The agent rewrote the first sentence to answer the page's question directly, with PrintStash as the subject, and moved context and caveats further down. These changes showed up in answers within 1 or 2 days.

  As a specific example, the roundup now opens with "The best self-hosted 3D print file manager is PrintStash". After that rewrite on August 17, Google AI Overviews cited it two days later, and AI Overviews, AI Mode and Gemini went from zero to between 9.5% and 23.5% share of voice on that question within 30 days.

• Fixing wrong sentences on PrintStash's own site. When answers repeated something wrong about PrintStash, the agent traced the claim back to the sentence engines were quoting on PrintStash's own pages and corrected it there. 11 of 14 of these fixes improved.

• Shipping content with each product release. Product releases became one of the most important signals for what to write and update, since each one brings new features to cover and existing pages that need correcting.

• Measuring well after the ship date. The first days after a change ships say little, because engines might take several days to pick it up, and some changes classified as "no change" at day 7 had improved by day 30.

• Getting listed where engines already look. Adding your product to popular, established listings borrows their authority and audience, and engines already cite them. PrintStash's listing in the Unraid app store, a popular home-server platform, was the most cited source across our prompts in early August.

What didn't

• Cold outreach. Pitching editors to add PrintStash to their articles rarely worked, because most never replied.

• Release notes as content. Release posts are written around a version, not around a question people ask, and they didn't earn citations. None of the three release posts we measured improved, and none of 24 release notes had been cited.

• Counting citations as wins. Engines often cite a PrintStash page without naming PrintStash, so a page can collect citations while the answer never mentions the brand. Making PrintStash the subject of each page's first sentence is what gets it named.

  In October, on the 19 prompts where PrintStash was weakest, Perplexity cited PrintStash in 99.4% of answers but named it in only 16.4%.

• Search Console impressions as a KPI. Omnia's daily monitoring shows up in Search Console as impressions, so impressions overstate how many people actually saw the site.

  In October, 52% to 56% of the site's impressions came from our own monitored prompts, with zero clicks.

How each engine behaved

This is what we saw working on PrintStash, one new product in one niche, so read it as experience and heuristics rather than hard rules. Each engine responded differently, and the same change could show up in days on one engine and weeks on another.

• ChatGPT and Perplexity cared most about content. Matching a page to the exact question a prompt asks was enough to get cited, even though printstash.org was brand new and had no domain authority.

  Perplexity picked up new pages fastest but often cited them without naming PrintStash. ChatGPT, on the other hand, usually recommended PrintStash by name whenever it cited one of its pages.

• Google AI Overviews, AI Mode and Gemini valued authority. For a new site, these were the hardest to win. Domain authority seemed to matter a lot, and in our experience ranking well in Google's organic results was a requirement for getting cited there.

  AI Overviews was the slowest to respond, and when it did, it was mostly to updates on pages it already cited before rather than to new pages. Gemini only started citing a new page regularly about 19 days after it shipped.

• Copilot favored older pages. It mentioned PrintStash in more answers than any other engine, 80.1% from September 26 to October 26, but when it cited a source it tended to pick older printstash.org pages over the new page written specifically for that question.

  That made outdated sentences on old pages costly, since Copilot kept repeating them until those pages were corrected. It's why the weekly Blog curator task, which checks every claim on the blog against the latest release, became critical for this engine.

Honest pages, and a product backlog Omnia prepares for you

Many of PrintStash's pages state what PrintStash doesn't do. Among all the posts on printstash.org's blog, 63 include at least one sentence like "PrintStash does not...", and comparison pages acknowledge where competitors win. The roundup on that site even opens by telling readers who wrote it:

"I work on PrintStash, so read the criteria and the per-tool sources below rather than taking the verdict on trust."

We learned this from the data and turned it into a rule. In our experience, when a product's pages don't state its downsides, AI engines sometimes fabricate them, and the made-up downsides can be worse than the real ones. Stating the real limits explicitly kept AI answers aligned with reality. After the homepage listed what PrintStash is not, all six engines started repeating those limits word for word instead of inventing downsides of their own.

However, leading with weaknesses backfired. What worked was opening with the job PrintStash does well and keeping the limits further down. For example, the PrintStash vs OctoPrint comparison opens with "OctoPrint runs a printer. PrintStash organizes the models, G-code revisions, and print history around it", and covers what OctoPrint does better further down in the article.

Honest pages have a second effect: AI answers show which of your weaknesses are actually hurting you, and how often they come up against competitors. Omnia's sentiment analysis breaks answers down by feature, showing which PrintStash features engines endorse and which they undermine.

Which PrintStash features AI answers endorse and undermine

Feature	Endorsed	Undermined
Printer integration	1,404	132
G-code revision tracking	1,242	8
Slicer metadata extraction	1,185	5
Print history tracking	736	5
File management	708	13
Klipper and Moonraker	575	5
Content-hash deduplication	515	2
File format support	393	240
Printer control and monitoring	89	348
Plugin ecosystem	2	327
Printer control	105	212
Maturity and stability	1	150
Project maturity	2	110
Setup complexity	10	107

Ranking the undermined features by how often they come up gives you a prioritized list of gaps, so the most mentioned ones get fixed first and the work lands where it has the most impact. That list can go straight into your product backlog to be triaged.

Some of the weaknesses in the chart are deliberate: PrintStash doesn't control printers and has no plugin system; the pages say so and point to OctoPrint, Mainsail or Fluidd for that job. The rest are real gaps, and those go to the PrintStash product backlog as feature requests Omnia Agent writes from the evidence in the answers. When the feature ships, Omnia Agent updates every page that mentioned the limitation, closing the loop. Three examples:

• Multi-printer dispatch. The guide to sending G-code to multiple printers admitted there was no way to send one file to several printers, or to stop a file prepared for one plastic from going to a printer loaded with another. The agent filed a feature request with the AI-answer data attached on August 17, PrintStash 0.12.0 shipped both two days later, and the pages were corrected the day after.
• Duplicate detection. Answers favored a competitor for finding similar models by shape, and PrintStash's own duplicate-detection guide admitted it only spotted exact copies. Similar Models shipped in 0.14.0 three weeks after the feature request. After the guide's update, answers naming PrintStash's Similar Models feature went from 6% to 48%, and the share of duplicate-detection mentions undermining PrintStash fell from 9.5% to 1.4%.
• Setup complexity. Copilot kept warning that PrintStash is hard to set up, partly by quoting a line in PrintStash's own Docker Compose comparison, which PrintStash 0.14.0 improved. That signal is what pushed PrintStash to make its setup much simpler, clearly improving the product's experience.

Over time the loop improves the brand's AI visibility and the product at the same time. If you are building a product, branded prompts show you which features engines think you're missing compared with your competitors.

Omnia now does most of what my custom workflow did

Several things I had to build by hand are now available to every Omnia customer out of the box. Dogfooding like this didn't set our roadmap on its own, but it gave us very good feedback on the foundations of these features:

• My "Generate Projects" task used to be my main source of work to do. Project Suggestions took over that part: Omnia Agent proposes up to three scoped Projects with their evidence every day. My custom task now only supports it when I need more Projects.
• My Notion board did the job Omnia Projects does now, and by October 26 my Omnia workspace had 170 Projects, 137 of them done and 33 discarded.
• The measurement step is still a custom workflow on top of the agent, because Projects record what shipped, but they stop there.

Beyond AI answers: clicks and stars

AI visibility only matters if it brings people to the product. PrintStash is open source and has no revenue channel yet, so there's no revenue to show, but two numbers that measure people finding and using it grew in the same period. I can't prove a direct link to this work, since PrintStash also grew through its own releases and word of mouth, and I don't think AI answers explain them on their own.

The work compounded. Omnia Agent's research kept finding places outside AI answers where PrintStash should be known, such as self-hosted app stores, tool directories and other people's roundups, so the brand became known in many more places than AI answers alone. For a new product, knowing where and how to get known is often the hardest part, because some channels are hard to find without help, and that turned out to be one of the most valuable side effects of the whole process.

• Average daily clicks from Google Search grew about 9 times. They went from 2.5 in the first two weeks of monitoring (July 8 to 21) to 22.8 from September 26 to October 26. 595 of the 1,556 clicks in the whole period came from people searching directly for printstash by name.
• GitHub stars grew from 28 to 230. The PrintStash repository had 28 stars on June 19, before monitoring started, and 230 on October 28.

Google Search clicks to printstash.org grew about 9 times

Hover, or focus the chart and use the arrow keys, to read any day.

• Clicks

See the data

Date	Clicks
Jul 8	0
Jul 15	2.1
Jul 22	3.9
Jul 29	5.4
Aug 5	3.9
Aug 12	4.3
Aug 19	5.9
Aug 26	6.6
Sep 2	8
Sep 9	11.3
Sep 16	17.4
Sep 23	43.7
Sep 30	29.4
Oct 7	24.4
Oct 14	21.1
Oct 21	21.1
Oct 26	18.3

Six things I'd do again

• Track the questions your buyers ask when they don't know your brand yet. 77 non-branded and 26 branded prompts per language were enough in our case.

  That gives you enough high-level data to clear the fog around your market and find plenty of levers to pull.

• When you target a specific prompt, answer it on its own page, with the answer and your brand name in the first sentence.

• Be honest about your limits on every page and acknowledge where competitors are better, but lead with what you do well.

• Read branded sentiment as a product backlog: fix the product, then update the page.

• Treat every change as a hypothesis: write down which prompts it should move and why, then validate it by measuring more than once over time (for example, at 7, 14 and 30 days) from the real ship date against a stored baseline, and use what you learn as feedback for the next round of the loop.

• Let the agent make changes directly to your website under a short list of hard rules, then review its work regularly.

  This works best when your website lives in a git repository, where an agent can change any page and ship it like a developer would. Visual low-code CMS platforms were born when code was a hard skill most marketers didn't have, and agents waste time and tokens just finding their way around them, which is why we moved Omnia's website from Webflow to Astro.