No-Code Automation Statistics 2026: 40 Verified, 25 Rejected

The no-code market-size number you keep reading is from 2021.

$44.5 billion by 2026 traces to a Gartner press release published in 2021. The 4-to-1 citizen-developer ratio traces to Gartner in 2021. The 75% of new applications line, same release family. The 16.2 million citizen developers figure has no origin document at all, in any year, from any publisher.

So this page carries no low-code market size. None.

What it carries: 40 numbered statistics that survived a source check on 15 September 2026, and four predictions labelled A to D. Every figure names its publisher, its publication date, the method the publisher disclosed, and a link to the document itself. Where a figure comes from a company with money riding on the answer, the line says so in plain words.

25 claims did not survive. They are listed near the bottom with the reason each one was dropped. That table is the part of this page I expect other writers to take. For the adjacent ground on agent adoption specifically, AI agent adoption statistics carries a separate set of figures and repeats none of these.

Quick answer

There is no citable 2025 or 2026 figure for the size of the no-code and low-code market, and no citable count of citizen developers. Both themes came back empty. The largest verified adjacent number is a forecast. Gartner forecasts that $234 billion of enterprise application software spend is at risk from agentic arbitrage through 2030, about 20% of enterprise SaaS spending, in a release dated 1 July 2026. The most solid measured statistic in the pack is GitHub's: 693,867 public repositories importing an LLM SDK created in the 12 months to 31 August 2025, up 178% year on year. Forty statistics follow, each with publisher, date, method and confidence, and then the list of what had to be thrown away.

What the build-it-yourself layer is worth now

Start with the only large 2026 figure that survives a source check. Gartner forecasts that $234 billion of enterprise application software spend is at risk from agentic arbitrage through 2030, and expects that to account for about 20% of enterprise SaaS spending by then.

Gartner, press release dated 1 July 2026, forecast attributed to George Brocklehurst, canonical release. Primary, though Gartner blocks automated fetch, so both figures were confirmed against CIO Dive's report of 6 July 2026.

That is a forecast of displaced spend. Quote it as the software budget agents are expected to eat and it holds. Quote it as the no-code market and it falls apart.

The private money moved first. Four rounds, all inside sixteen months, all announced by the companies raising them:

• n8n, $5.2 billion. After SAP's strategic investment of about $60 million at SAP Sapphire, roughly double the prior $2.5 billion. n8n via PR Newswire, 12 May 2026, announcement, corroborated by Bloomberg. Vendor-interested, because the company announced its own valuation.
• Lovable, $13.3 billion. On a $400 million Series C led by Menlo Ventures and the Scaleup Europe Fund, with $500 million ARR as of June 2026 against a $6.6 billion valuation in December 2025, 60 million projects hosted since launch and 1 million new projects a week. TechCrunch, 12 August 2026, report, which makes the round secondary; the other three figures are vendor-interested and self-reported. Sixty million projects is a lifetime count of what was started, and it says nothing about what still runs.
• Replit, $400 million at a $9 billion valuation. Series D led by Georgian, tripling the valuation in six months, alongside a claim of being on track for $1 billion ARR by end of year and more than 500,000 professionals on the platform. Replit, 11 March 2026, announcement, corroborated by TechCrunch the same day, so the round is vendor primary and corroborated. The ARR line is a claim about the future, and the professional count is self-reported with no published definition.

Against those valuations sits the incumbent's scale. Zapier's press kit claims 3.4 million businesses and more than 9,000 apps in its directory (zapier.com/press). Vendor-interested and undated: the page prints no date, so both figures are recorded here as of September 2026 and nothing stronger.

An undated vendor number is the most quoted and least checkable object in this category. It gets copied for years with no way to tell when it stopped being true. The AI agent pricing index covers what these companies charge, and it dates every figure it prints.

Who actually builds now

The citizen-developer population is unknowable from public sources, so this theme replaces it with behaviour that can be counted.

Stack Overflow's 2025 Developer Survey, about 49,000 respondents across 177 countries, self-selected: 84% of developers use or plan to use AI tools in their development process, and 51% of professional developers use them daily (survey.stackoverflow.co/2025/ai, published 2025). Primary, with the standing caveat that a self-selected sample skews toward developers engaged enough to answer.

The agent numbers in the same survey are lower and much sharper. 30.9% use AI agents daily, weekly or monthly, and 37.9% do not plan to use agents at all. Same survey, same link, and primary. This is the only figure in the pack that measures refusal, and it is the one to hold up next to every vendor adoption percentage further down.

GitHub counts artefacts. In the year to 31 August 2025 the platform recorded more than 180 million developer accounts and 36 million new developers, up 23% year on year, about one a second (GitHub Octoverse 2025, 28 October 2025, platform telemetry). Primary, but it counts accounts and not people doing work.

The strongest single line in the whole pack is the next one. 1.1 million public repositories import an LLM SDK, and 693,867 of them were created in those 12 months, up 178% year on year. Same release, same telemetry window. Primary: a count of repositories, with nobody asked to self-report.

Read it as the size of the thing no-code competes against. Nearly 700,000 new repositories of hand-built agent plumbing in a single year.

GitHub also reports nearly 80% of new developers using Copilot in their first week. Same release. Primary telemetry and vendor-interested at the same time, since GitHub is measuring uptake of its own product.

Microsoft narrows the same question to the workforce. 16%, or 3,233 of 20,000 AI-using knowledge workers qualify as what Microsoft calls Frontier Professionals: people routinely running agents on multi-step workflows (Microsoft 2026 Work Trend Index, 5 May 2026, fielded 18 February to 7 April 2026 by Edelman Data x Intelligence across 10 countries), which makes it vendor-interested with the method disclosed. The sample is already AI-using, so 16% of that group is a far smaller share of any general workforce.

Forrester puts the work itself at 48% coding, 47% testing and 33% development insights as the top AI use cases among software teams, and predicts a 20% drop in computer science degree enrolment (Forrester, Predictions 2026: Software Development, 4 November 2025), primary analyst work for the use cases, with the sample size not restated publicly. The enrolment figure is a prediction. Nobody has counted it.

Agents inside the automation stack

Only about a quarter of enterprises hand-code their agents. That single fact is the bridge from no-code automation into this category, and it comes from Zapier.

The method first, because it decides how much weight the rest carries. 525 surveys run by Centiment, fielded 23 to 27 October 2025, respondents were US C-suite executives, owners and partners at companies with 1,000 or more employees, margin of error about 4% (Zapier, State of agentic AI adoption, 15 December 2025). Zapier sells agent tooling, so call it vendor-interested. The method is disclosed, which is more than most vendor surveys offer.

• 72% of enterprises use or are testing AI agents.
• 40% run multiple agents in production, and 32% are in pilot or testing.
• 84% plan to increase AI agent investment in 2026. That is intent, not spend.
• By function: customer support 49%, operations 47%, engineering 35%, marketing 31%, sales 26%, finance 24%.
• By build route: 26% custom coding, against 53% cloud providers, 48% open-source tools, 48% enterprise AI platforms and 46% orchestration frameworks. Multi-select, so the shares do not sum to 100.
• By oversight: 38% use human-in-the-loop as their main management approach, and 20% run agents autonomously with minimal oversight.

Set the 26% next to Stack Overflow's 30.9% and the shape gets clearer. Most enterprises buy the agent layer. A minority builds it. That is the same trade no-code always offered, arriving again under a different name, and AI agent vs workflow automation is where I draw the line between the two.

Two telemetry numbers follow, and both need a label. Microsoft reports 15x year-on-year growth in active agents across Microsoft 365, and 18x in large enterprises (Microsoft 2026 Work Trend Index, 5 May 2026, aggregated Microsoft 365 telemetry from March 2025 to March 2026). Vendor telemetry from a small base, and Microsoft's own product at that. A 15x multiple off a low starting count is an easy number to produce and a hard one to interpret.

Salesforce reports nearly 3x growth in activated agents between February 2025 and April 2026 (Salesforce Agentic Enterprise Index, 7 August 2026), vendor-interested and survivorship-biased. The cohort is defined as customers who activated agents every single month, so anyone who gave up is excluded by construction. Cohort size is not published.

Survivorship bias is the whole story here. A cohort selected on continued use will always show growth, improvement and rising sophistication, because the failures were filtered out before the chart was drawn.

Time to build, and what it returns

Every vendor time-saved claim in this section has the same counterweight, so the counterweight goes first. Stack Overflow's 2025 survey found 66% of developers hitting AI output that is almost right but not quite, and 45.2% saying debugging AI-generated code takes longer. About 49,000 respondents, published 2025, which makes it primary and independent of anyone selling the time saving.

Hold that pair next to each claim below.

Salesforce's cohort shows a 53% reduction in average agent creation time, with agents growing from 2 skills to 6 across 2025 (Salesforce Agentic Enterprise Index, 7 August 2026, product telemetry, cohort size undisclosed). Vendor telemetry again, on the same survivorship-biased cohort. The skills figure measures configuration. It says nothing about outcomes.

The same index reports 7 of 10 service chats resolved without a human, with the escalation rate steady at 32%. Confidence: vendor-interested, and those two figures sit in obvious tension. Seven in ten resolved, roughly a third still escalated. Ask about the arithmetic before repeating either half, and quote them together when you do.

The ROI figure in circulation is modelled, not measured. Forrester Consulting, commissioned by Microsoft, projects a three-year ROI of 106% low, 216% mid and 314% high, with NPV of $25.7 million, $52.6 million and $76.4 million against $24.4 million of costs for Copilot Studio deployments (Forrester Total Economic Impact, September 2025, built from 13 customer interviews plus 400 survey respondents combined into a composite organisation), which makes it vendor-commissioned and modelled. The word projected belongs in every sentence that uses it.

Microsoft's self-reported outcomes: 58% of people using AI at work, rising to 80% among Frontier Professionals, say they produced work they could not have produced a year earlier, and 66% report more time on high-value work (Microsoft 2026 Work Trend Index, 5 May 2026), vendor-interested and self-reported, with no external measure of the work itself.

Two different 66% figures, pointing opposite ways. One asks people how they feel about their time. The other asks developers what the output cost them to fix. Both are real. Only the second is independent of a company selling the tool.

If you are choosing between platforms on the strength of numbers like these, the workflow automation tools comparison deals in published prices and billing units. Those are checkable line by line. A modelled ROI is a projection.

What breaks

No public dataset measures no-code project abandonment. So the honest answer to whether these projects fail comes from adjacent AI research, and it is labelled as adjacent.

S&P Global Market Intelligence found 42% of businesses scrapping most of their AI initiatives, up from 17% a year earlier, with the average organisation abandoning 46% of AI proofs of concept before production. Survey of more than 1,000 respondents across North America and Europe, published March 2025, reached through CIO Dive's reporting, so it is primary research arriving through a Tier 3 outlet. Both halves belong in the same sentence.

Gartner's governance prediction: 40% of enterprises will demote or decommission autonomous agents by 2027 because of governance gaps found after production incidents (Gartner, 26 May 2026, prediction attributed to Shiva Varma, verified against CIO's report of 29 May 2026). Confidence: primary, and a prediction. This is a different figure from Gartner's better-known line about agentic projects being canceled, and the two get merged constantly.

The oversight numbers explain the mechanism. Zapier surveyed 518 executives at companies that already have an AI governance policy and found 85% keep some human approval and 15% grant full unsupervised autonomy, while 38% have already suffered negative consequences from unsupervised AI, of whom 39% reported delayed or lost revenue and 36% legal, regulatory or compliance consequences (Zapier, no publication date printed, read 15 September 2026). Vendor-interested, undated, and screened. The population already has governance in place, so the figure flatters the market.

Nearly two in five of the most governed companies have already been burned. That is the line I would carry out of this page.

The structural shift, in four predictions

Four forward-looking figures. All labelled predictions. None of them measurements, and none of them should be quoted as if they were.

• A. 60% by 2029, up from 15% in 2026: organisations adopting smaller software engineering teams at scale, typically four to five people. Gartner, 7 July 2026, verified via People Matters on 8 July 2026. Confidence: primary, and a prediction.
• B. At least 50% by 2029: knowledge workers who will develop new skills to work with, govern or create AI agents on demand. Gartner, 21 October 2025. Confidence: primary but snippet-verified only. Treat it as supporting evidence and keep it out of a headline.
• C. Time to hire developers will double. Forrester, 4 November 2025. Confidence: primary, and a prediction published with no baseline figure beside it.
• D. Embedding by Q3 2026: SAP's stated plan to put n8n's visual canvas and more than 1,000 integrations inside Joule Studio. n8n and SAP, 12 May 2026. Confidence: vendor-interested and directional. A stated plan, with the shipping still ahead of it.

Prediction D is the one worth watching, because it is the only one with a date this quarter and a checkable outcome. Either the canvas is inside Joule Studio by the end of September 2026 or it is not. The other three resolve in 2029, which is a long time to quote a number nobody can mark wrong.

A and C describe one world from two sides: teams of four to five people, and twice as long to hire each of them. If both land, the case for buying a finished agent makes itself. No-code AI agent platforms covers that side of the field.

The numbers that do not exist

Five themes came back empty or thin. Saying so out loud is the point of this page.

Empty: no verified data exists

1. No-code and low-code market size, published 2025 or 2026. Empty. The Gartner document that would settle it (7146430) is paywalled. The $44.5 billion by 2026 line comes from a 2021 release. Forrester's $50 billion by 2028 blog post is dated 29 January 2024, outside a two-year window. This page therefore carries no low-code market size at all.
2. Citizen-developer population and ratios. Empty. The 4-to-1 ratio, the 80% built outside IT and the 75% of new applications figures all trace to Gartner material from 2021 and 2022. The 16.2 million citizen developers figure has no origin document. The measured developer behaviour in who actually builds now replaces it.
3. No-code app abandonment rates. Empty. The only survey behind the circulating failure numbers is a CIMI study from 2018. What breaks reframes the question using AI project data, and says plainly that no-code-specific failure data is not public.

Thin: some data exists, none of it clean

1. Time to build, independent of vendors. Thin. The substantial study is Forrester's Power Apps Total Economic Impact from July 2024, which is both outside the window and commissioned. Every build-time figure on this page comes from a company selling the tool.
2. Governance of citizen-built applications. Thin. The widely quoted line that 78% of organisations have formal citizen-developer governance has no Gartner source behind it. The Zapier figures above cover AI governance generally, on a population screened for already having a policy.

The 25 claims I dropped

Each of these turns up in no-code and low-code statistics round-ups. None survived a check against an original document published in the last two years.

Dropped on 15 September 2026. A claim being dropped here does not mean it is false. It means I could not trace it to an original document published within two years, or the original contradicts the figure being quoted, or the publisher's own page carries a different number. The last two rows are on hold: both may be usable once verified by hand, with their caveats attached.

Every figure on one table

44 figures, one row per highlighted number on this page; the # links to the sentence it sits in and the publisher links to the original document where the link is recorded. Confidence is the CSV label as written: primary means the publisher collected the data; secondary means a credible re-report or model; vendor-interested means the publisher sells something the figure flatters; forecast means a prediction, not a measurement; unverified means the original blocked our read and the figure comes from a named re-report. The same rows are in data.csv (CC BY 4.0).

Use this data

Take it. That is why it is here.

The CSV carries eight columns: id, figure, context, publisher, date, method, url, confidence. Every highlighted figure on this page has one matching row, 40 numbered and four lettered, 44 in total.

Two requests if you use it. Quote the confidence note alongside the figure, because vendor-interested is doing real work in about half these rows. And when all you need is the number, link to the publisher's document.

Corrections are welcome, and early ones are best. If a source has published a newer edition, or a figure here reads differently on the original page, the next quarterly pass picks it up. Corrections: support@gravity.fast, with the source URL and the date you read it.

FAQ

Sources