AI has sharply reduced the cost of producing software. But markets are such that cheaper production doesn’t automatically turn into higher sales. As a result, consumption won’t grow — competition will. And lower costs in IT teams and IT companies aren’t guaranteed by themselves: savings simply flow into sales channels, where competition gets more expensive, and those who can sell in tough conditions are the ones valued.
This article is an attempt to look around and soberly assess the consequences. Yes, development has accelerated. But was it the main bottleneck before AI? Most often — no.
1) AI accelerates development. That’s a fact. And that’s the expectations trap
AI really does make engineers more productive. In a controlled experiment with GitHub Copilot, developers completed the task about 55.8% faster.
Many infer from this: “now you can build a product quickly — so you can make money quickly.” But build speed is only one variable.
If it’s become easier to “make,” that doesn’t mean it’s become easier to “sell.” On the contrary: if “making” gets cheaper for everyone, there will be more competitors.
2) Development was rarely the main bottleneck. Sales were — specifically, profitable sales
Sales alone guarantee nothing. What matters is profitable (margin-positive) sales: when the cost to acquire a customer is lower than what that customer will bring over their lifetime.
A simple formula:
- CAC — how much it costs to acquire a customer (marketing, sales, lead handling, pilots, negotiations).
- LTV — how much the customer brings over their whole lifetime (revenue minus support/infrastructure/ongoing service).
If CAC ≥ LTV, you’re not growing — you’re slowly burning money.
Even before AI, most products died not because “they couldn’t write the code,” but because they:
- didn’t find a working acquisition channel;
- missed positioning;
- couldn’t prove value and make sales pay back.
AI doesn’t cancel this part — it makes it worse.
3) The technical barrier has “melted” → competition in sales channels will intensify
If earlier building a product was expensive and slow, supply was constrained by the team’s labor.
Now supply is constrained less. Which means:
- there will be more products;
- there will be far more similar products;
- noise in channels will increase;
- and buyers’ attention will remain the same.
Herbert Simon has a good line: “a wealth of information creates a poverty of attention.” AI increases the amount of information. But it doesn’t add attention.
Hence a simple effect: channels will get crowded.
4) “A lot will be built. Roughly the same amount will be sold” — why this isn’t cynicism, but economics
Important: the advertising market is growing. But it’s not growing by multiples.
According to forecasts from major advertising-market players, global ad budgets are on the order of a trillion dollars a year and grow by percentages, not exponentially.
This means:
- the overall “attention pie” expands slowly;
- meanwhile the number of advertisers (including new AI products) grows;
- therefore competition in auctions grows faster than the market itself.
Hence the unpleasant logic: for your new product to be bought at scale,
- someone has to stop buying the old one,
- someone has to be pushed out of the market,
- or you have to occupy a new niche where competition hasn’t yet turned into an auction.
5) The paradox: AI helps salespeople, but barely helps win the fight for attention
AI will of course be useful in sales operations:
- emails, scripts, proposals;
- preparing materials;
- initial lead qualification;
- market and competitor analysis.
But it’s a symmetric weapon. Your competitors will do the same.
And the main thing — access to attention — AI barely expands.
Sales channels are competition for human attention. And attention doesn’t grow just because there are more products.
6) Who will win from the AI race in application software
6.1. Distribution platforms and advertising ecosystems
When more products and advertisers appear, the winners are those who own:
- the search entry point,
- social feeds,
- the ad auction,
- analytics and targeting.
In Russia this is especially visible: according to StatCounter for January 2026, Yandex’s search share is about 70.5%, Google’s about 27.4%.
If the market is “flooded” by a wave of new AI apps, the ad auction becomes more expensive. Which means platform rents grow.
6.2. Incumbents with trust and adoption
The cheaper it is to create “yet another product,” the more valuable the things become that are hard to copy:
- trust,
- adoption,
- integrations,
- support,
- accountability,
- user habit,
- compliance and closed/perimeter environments.
That’s why AI simultaneously:
- helps newcomers assemble prototypes faster,
- and strengthens those who are already embedded in customers’ processes.
7) Natural barriers: where AI “bogs down” — and why that’s an opportunity
It’s worth recalling the plot of The War of the Worlds: the invasion seemed unstoppable and superior to humans in every way, but what proved decisive were not “countermeasures” — it was the constraints of the environment, those very natural barriers.
AI is similar: it won’t stop or retreat. But it will keep running into barriers. And wherever it stalls, demand emerges for new roles and services.
Below are five barriers that are already clearly visible.
I. Energy: the pilot is cheap, industrial operation becomes part of the business model
In a demo everything looks rosy. In production mode — 24/7, SLAs, peaks, incidents — the cost per response and the infrastructure become part of your unit economics.
Many companies hit the wall precisely at the transition “pilot → rollout.” According to an IDC study (in partnership with Lenovo), 88% of PoC-level initiatives don’t reach wide deployment; out of 33 pilots, 4 make it into production.
II. Chips and infrastructure: scaling is not only an idea, it’s supply
Hardware is produced by a limited number of fabs; supply chains depend on geopolitics and logistics; company upgrades depend on annual budgets and KPIs.
So “scaling AI” isn’t just “hire developers.” It’s operating infrastructure in a world with external constraints.
III. Trust and security: AI errors are confident, and attacks are cheap
AI can be confidently wrong. Plus there are attacks through content and instructions. Plus leaks. Plus dependence on a black box.
As long as there is no mechanism that reliably eliminates output fallibility, trust is built not on “belief in the model,” but on guardrails, control, accountability, and evidence.
IV. Data: real business context is closed, messy, and distributed
In business, what matters isn’t “approximate solutions,” but specific ones. And the data:
- lives in different systems,
- has exceptions,
- requires interpretation,
- and is often not ready for use.
Bad data and incomplete context turn any “smart” answers into a beautiful but dangerous approximation.
V. Matter: a mistake in the physical world is an incident
In chat, you can clarify a mistake. In logistics, manufacturing, medicine, and finance, a mistake becomes an incident. That’s why AI’s penetration into the physical loop will be slower — and the price of trust will remain high.
Conclusion: AI won’t stop. But the “invasion” will be uneven. And durable markets emerge precisely where the cost of mistakes is high and where there are infrastructure and organizational barriers.
8) Four groups of players — and what they should do
1) AI startups: learn not to “build,” but to sell with payback
If earlier a startup could burn the investment without finishing the product, now it more often dies without building sales.
The practical advice sounds unexpected:
- understand marketing and positioning more deeply than technology;
- don’t reinvent the wheel in go-to-market;
- bring in experienced product marketers earlier than “too late.”
Because now money doesn’t burn so much on development. Now it burns primarily on attempts to “find a channel,” while competitors ship a product a week too.
2) Marketers and “digital people”: demand will grow, but in a different format
Opportunity: small teams will need help with “packaging and funnel” — fast, pragmatic, without bureaucracy.
But the format will more often be not “one person full-time,” but:
- a portfolio of projects,
- short iterations,
- hypothesis-driven work,
- transparent metrics,
- pay for outcome.
In a crisis, this can become a real way to survive and grow.
3) Indie hackers: you can live away from the bloody ocean
Indie hackers can ignore many problems: their projects don’t break the bank; they don’t need scale to earn.
But they rarely set the tone for the market:
- products live in short cycles,
- often don’t build trust,
- and don’t sustain industrial-grade support.
4) Those who don’t want the bloody ocean: look for niches with natural barriers
If you don’t want to participate in the auction of “who’s louder and cheaper,” look for areas where AI hits barriers:
- closed/perimeter environments,
- security,
- data,
- adoption and integrations,
- operations and accountability,
- the physical loop,
- cost of mistakes.
In these areas, competition is harder. But margins hold longer and trust builds faster.
9) Founder checklist: how not to die in the AI race
If you’re building an AI product, ask yourself five questions:
- What acquisition channel do we have that doesn’t turn into an endless auction?
- How do we ensure sales pay back: what is CAC and what is the real LTV?
- What protects us from copying if “building” has become cheaper?
- Where do we run into natural barriers — and can we turn them into a competitive advantage?
- What about trust: support, adoption, accountability, security?
If there are no answers to these questions, AI will accelerate you… but not in the right direction.
If you need to get through this stretch faster than competitors, we can plug in as an external team and in 4–6 weeks turn sales into a manageable system: find 3–5 bottlenecks, package the offer, build outbound access to decision makers and a “lead → contract” funnel, put order into CRM and analytics, provide a realistic forecast, and establish working standards for the marketing and sales team.
Conclusion: AI shifts the bottleneck in the value chain, but doesn’t remove constraints
AI makes engineers stronger and speeds up software production. But it doesn’t create demand and doesn’t expand attention.
AI doesn’t change the main thing:
- selling with margin is harder than writing code;
- mistakes didn’t get cheaper — they often got more expensive;
- there will be more competitors, but not more channels.
In this race, the survivors won’t be those who loudly promise “we’ll build it in a week,” but those who:
- can calculate acquisition unit economics,
- build trust and accountability,
- understand where AI’s “invasion” bogs down,
- and turn those barriers into margin protection.
Links and sources
- Experiment on speeding up development with GitHub Copilot (55.8% faster):
https://arxiv.org/abs/2302.06590 - WPP Media forecast: global advertising ~$1.08T (2025), growth ~6%:
https://www.reuters.com/business/media-telecom/wpp-media-cuts-2025-global-advertising-revenue-growth-forecast-6-trade-concerns-2025-06-09/ - WPP Media forecast update (official page):
https://www.wppmedia.com/news/tyny-midyear-2025 - Yandex/Google search shares in Russia (Jan 2026):
https://gs.statcounter.com/search-engine-market-share/all/russian-federation - Dominance of the advertising “triad” (55.8% outside China):
https://www.warc.com/content/feed/global-ad-growth-forecasts-upgraded-on-social-media-windfall/en-GB/10987 - IDC + Lenovo: 88% of AI PoCs don’t reach production (33 → 4):
https://www.cio.com/article/3850763/88-of-ai-pilots-fail-to-reach-production-but-thats-not-all-on-it.html - S&P Global: share of companies abandoning most AI initiatives rose from 17% to 42%:
https://www.spglobal.com/market-intelligence/en/news-insights/research/ai-experiences-rapid-adoption-but-with-mixed-outcomes-highlights-from-vote-ai-machine-learning - SaaS benchmark on rising blended CAC ratio (1.32 → 1.61):
https://joinpavilion.com/hubfs/2024%20B2B%20SaaS%20Performance%20Metrics%20Benchmarks%20Report.pdf - “Attention paradox” (Simon, Oxford Reference):
https://www.oxfordreference.com/display/10.1093/acref/9780191843730.001.0001/q-oro-ed5-00019845