More options than anyone can look at

The short version

• Project management software is multiplying faster every year. Listings on G2, which describes itself as the largest software review site, grew 5%, then 9%, then 30% year over year, from mid-2023 to mid-2026.
• Upstream it is faster still. New open-source project management projects went from 413 created in 2023 to 3,883 in 2026, with the year not yet over.
• Behind both is the cost of building software, collapsing in plain sight. Apple reviewed more than 9.1 million app submissions in 2025, up from 6.1 million in 2022. In Apple's words, "powerful AI development tools drive a surge in app submissions."
• The same cost curves are bending in the physical world. Battery packs are about 90% cheaper than in 2010, and solar modules about 500 times cheaper than in 1975.
• Buyers are moving the other way. The average software buyer now actively engages 2.5 vendors, down from 3.2 a year earlier.

1. Building keeps getting cheaper

Every technology made at scale gets cheaper to make.

These curves are not smooth. Battery and solar prices both rose in 2022, and genome costs have fallen far more slowly since 2015. But the direction over decades is not in doubt, and it is the direction that matters here: when making something gets cheaper, more people make it.

2. Software is where you can see it first

AI has made software the fastest-moving case.

• The price of intelligence. Running a language model at a fixed level of capability fell from about $60 to $0.06 per million tokens between late 2021 and late 2024, a thousandfold drop (a16z). Epoch AI measures declines of 9x to 900x per year, depending on the task.
• More software, sooner. GitHub counted 121 million new repositories in 2025, about twice its 2020 figure (Octoverse 2025). Apple reviewed more than 9.1 million app submissions in 2025, new apps and updates together, up from 6.1 million in 2022, and wrote that "powerful AI development tools drive a surge in app submissions" (Apple).
• Smaller teams. The share of new startups with a single founder rose from 23.7% in 2019 to 36.3% in the first half of 2025 (Carta).
• Faster to market. In 2025, 20% of startups formed through Stripe Atlas charged their first customer within 30 days, up from 8% in 2020 (Stripe).
• More companies. Americans filed 5.67 million applications to start a business in 2025, up from 3.50 million in 2019 (US Census Bureau, our annual sums of the monthly data).

The caveats belong here too. In a controlled 2025 study, experienced open-source developers took 19% longer with AI tools, while believing they were faster (METR). The surge in business applications began in 2020, before today's AI tools, and the applications most likely to become employers have been flat since 2021. Cheaper building does not make every builder faster. It lets more people build at all.

3. Project management, counted

Project management is a useful test because it was crowded already. Every directory we could compare grew, and the biggest one is growing faster each year.

Zapier's directory lists only apps that connect to Zapier, so it counts a narrower slice of the market than the others.

Capterra and GetApp stop at January 2025. After that, both sites changed what their category pages count, and their totals swing by hundreds within weeks.

G2 shows the shape most clearly, because it is one category page read about every quarter for three years:

G2 itself wrote in July 2026 that "there are 676 listings in the Project Management category on G2 today," which matches the pages. At the end of August 2026 the page jumped to more than 1,300 listings in under two weeks, so the comparison stops at 769 on August 18, 2026.

Directories trail the market: a product appears only once it has a website, a listing and usually some reviews. To see what is being built now, look upstream. New public GitHub repositories tagged project management:

Not every repository becomes a product, and part of the 2026 jump is tooling for managing AI agents, itself a new kind of project management. It is still the clearest view of where the next directory listings will come from.

4. It is not only software

The long record outside software points the same way.

Growth is not endless in every market. The average US supermarket carried about 47,000 items in 2010 and about 33,000 in 2025 (FMI), and the marketing technology landscape grew less than 1% this year. Markets consolidate, retailers prune shelves and directories curate. Those pauses are real, and they are local. The cost of creating the next option keeps falling.

5. The physical world is next

Software shows what happens when building becomes cheap. The early signs that physical products are heading the same way:

• Electric car models on sale worldwide went from about 450 in 2021 to nearly 785 in 2024, and the IEA expects more than 1,000 by 2026 (IEA 2022, IEA 2025).
• Industrial robot installations reached 542,000 in 2024, more than double ten years earlier (IFR). A humanoid robot now lists from $4,900 (Unitree).
• AI is starting to design physical things. Google DeepMind's GNoME predicted 380,000 stable crystal structures (DeepMind). AI-discovered drug candidates have passed Phase I trials at 80 to 90%, "substantially higher than historic industry averages" (BCG analysis, 2024), though the sample is small and covers only the earliest stage of testing. Both results are contested, and neither has yet reached the count of products people can buy.

This part is a forecast, not a measurement. The cost curves are the reason to make it: what software went through first, manufacturing, materials and medicine are beginning to go through now.

6. Attention does not scale

The number of options grows. The time and attention people give a decision does not.

• Software buyers start with 4.4 options, shortlist 3.5 and actively engage 2.5 vendors, down from 3.2 a year earlier (Gartner Digital Markets, 2025).
• 83% of buyers shortlist three products or fewer, 67% buy their first choice, and 79% knew the product before they started researching (TrustRadius 2026, summarized by its parent HG Insights).
• 59% of software buyers regret at least one purchase made in the past 18 months (Gartner Digital Markets, 2025).

Does more choice cause worse decisions? The research is more careful than the popular story. The famous jam study has not replicated reliably, and a 2010 meta-analysis of 63 conditions from 50 experiments found an average effect near zero (Scheibehenne, Greifeneder and Todd). A 2015 meta-analysis found that overload does appear reliably under four conditions: a complex set of options, a hard decision, unclear preferences and a wish to keep the effort low (Chernev, Böckenholt and Goodman). Choosing business software meets all four.

7. Who decides what you see first

When buyers look at only two or three options, the decision is mostly made by whatever put those options in front of them.

• In January 2026, G2 agreed to buy Capterra, Software Advice and GetApp from Gartner, putting four of the best-known software review sites under one owner (G2).
• Half of B2B software buyers (51%) now start their research with an AI chatbot more often than with Google (G2, 2026).

That second number changes who builds the shortlist. A page of search results shows ten links, and a buyer can see the ads for what they are and scroll past them. A chatbot answer names three or four products in a paragraph and does not show what it left out or why. The afternoon of reading that used to produce a shortlist now arrives in one reply, and whatever shaped that reply shaped the decision.

Search sells the top of the page. AI assistants tend to repeat whoever has published the most about themselves. Neither is built around the buyer's priorities.

What this means

More options are good news. They mean more people can build, and more of what gets built fits someone exactly. But more choice is not more freedom when the shortlist is set by whoever spent the most to be seen. As building gets cheaper, options stop being scarce. A fair way to choose among them becomes the scarce thing.

That is the work OpenDecision exists to do.

Method

Directory counts are the category totals each site printed on its own category page, read live and from archived copies in the Internet Archive's Wayback Machine. Where a site changed what its page counts, the comparison stops at the last point before the change. GitHub counts come from GitHub's search API and include only repositories that still exist. Every other figure comes from the publishing organization's own report, release or dataset, linked beside it. US Census annual totals are our sums of the Bureau's monthly data.