Building Got Free Indie software after coding agents

The middle of the software market is being squeezed shut.

The case: the indie hacker product is dying, and coding agents are the cause. Not because these products are bad, but because of where they sit. Too small to be defensible against a general agent, too large for one person to grow into something integrated, and now cheap enough that any customer can approximate it themselves. What follows is the strongest version of that case, laid out layer by layer, then the case against it: the constraint it measures was never the one doing the work.

01 The position, not the product
Pressure from above

Integrated scope you cannot staff

A general model with tools attached already covers thin capability. Above that sits real product surface: deep in one industry, or wide across a whole workflow. Both need a team, a roadmap, and years. One person with an agent does not get there.

The indie product

Viable band  ·  100% and closing

Valuable enough that someone will pay for it. Small enough that someone can reproduce it. That combination used to be the whole opportunity. The argument is that it is now the trap.

Pressure from below

An 80% version, coded in an afternoon

Your competitor is no longer another founder. It is your own customer, prompting their way to something that mostly works and costs them nothing per seat.

Fig. 1  ·  The seam thins from both faces at once Vertical axis: defensibility
02 The four load-bearing claims
CLAIM A

Not valuable enough to survive a general agent

If the product is mostly a prompt, a form, and a Stripe key, then a chat window with file access and two MCP servers does the same job without a subscription. The features were never the moat. The model brought the capability inside itself.

Sits in the blast radius
  • Chat with your PDFs
  • Meeting notes cleaner
  • Resume rewriter
  • SEO brief generator
  • Transcript summariser
  • Social post scheduler with AI captions
  • Scraper to spreadsheet
CLAIM B

Not enough capacity to build the thing that would survive

Defensibility now lives in scope. Vertical: one industry, all the way down through its billing, compliance, integrations, and edge cases. Horizontal: becoming the surface a whole team works inside all day. Both are staffing problems before they are engineering problems.

Agents compress the writing of code. They do not compress procurement, certification, support rotas, or the eight incumbent systems you must integrate with.

Out of reach for one person
  • Payroll in a regulated market
  • Clinical records integration
  • Two-sided marketplace liquidity
  • An ERP module
  • Bank-grade reconciliation
  • Anything with an auditor attached
CLAIM C

The floor rose, so the price anchor fell

Software that was worth buying because building it was annoying is now worth building because buying it is annoying. The moment a customer can generate a passable version, your price stops being compared to their time and starts being compared to nearly zero.

Bought yesterday, generated today
  • Form builders
  • Status pages
  • Link shorteners
  • Invoice generators
  • Internal dashboards
  • Browser extensions
  • Small CRMs
CLAIM D

Two categories are left standing

The labs, moving down out of the API and into the application layer they used to sell picks and shovels to. And venture funded startups, who can pay for enough engineers to stay ahead of what a prompt produces. The second group gets a reprieve rather than a position.

The two survivors
  • Labs shipping coding agents
  • Labs shipping spreadsheet and slide agents
  • Labs shipping browser agents
  • Funded editor and infrastructure companies
  • Anyone who can out-hire the prompt
03 Why those two, specifically
Survivor one

The labs come downstairs

They own the capability, the distribution, and the default surface people already have open. Every generic application feature is, to them, a rounding error in the roadmap.

Advantage: owns the layer beneath you

Survivor two

Capital buys the last 20%

The gap between something that demos and something a business runs on is reliability, migrations, permissions, support, and speed. That gap is paid for in headcount.

Advantage: temporary, and known to be

04 And then the ground itself goes

Applications

Value sits in the surface you open, the screens, the buttons, the seats you pay for.

A wrapper you speak to

Input becomes voice, the interface thins to a shell over the model.

Agents that just do it

No surface left to sell. The work happens without anyone opening anything.

On this reading, the indie hacker is not losing a market. The market is a temporary feature of a decade in which software still needed somewhere to be looked at.

The turn

The seam was never held open by code.

Every claim above measures the wrong constraint. Nobody's small software business was ever protected by the difficulty of writing it. It was protected, when it was protected at all, by being found and being believed. That is the axis the argument forgets, and it is the axis that does not deflate.

05 The rebuttal, as arithmetic
COUNTER A

Compute deflates. Attention does not.

Building is manufactured, so its price falls. Attention is fixed supply, so its price rises as more people bid for it with cheaper products. This is not a claim about difficulty, it is division. If building was most of the job and building goes to nearly free, then distribution becomes nearly all of the job. The bottleneck did not shift, it was simply the only bottleneck that was ever load-bearing.

Which is why the death notice does not follow from its own evidence. Falling build cost cannot kill a category whose failure rate was already governed by something else entirely.

c.2015
Build
Distribute
c.2021
Build
Distribute
Now
B
Distribute
Fig. 5  ·  Where a solo founder's week goes. Illustrative proportions, not measured data.
COUNTER B

Discovery moved from the index to inference

Search did not simply get worse. It failed twice at once. On the supply side, the marginal cost of content hit zero and the index filled with plausible filler, then answer panels took the click from whatever still ranked. On the demand side, the question "which tool should I use for this" increasingly gets asked of a model rather than a search box.

So the ranking game has been replaced by a citation game. What lands in a model's answer is what humans said about you in places models read: forum threads, video, repositories, actual coverage. Your own marketing copy is the one input that no longer counts.

The vaporware plus waitlist plus Stripe link routine died of the same disease. A landing page is now the cheapest artefact in existence, so it transmits no information about whether anything is behind it. When the artefact is free, evidence becomes the scarce good.

06 The channel ledger
Closed or closing

Rented reach

Organic search
Flooded from below, then answer panels took the click. Ranking now costs more than it returns.
Launch-day spikes
A day of traffic from people who ship things, not people who buy things.
Cold outbound
The tool that wrote your sequence wrote the four hundred others in that inbox.
Directories and listicles
Generated at scale, trusted by nobody, cited by nothing.
App store search
The same flooding problem, one layer down.
Still open

Accountable reach

An audience you own
A list, a community, a group chat of buyers. Nobody can reprice access to it.
Creators putting their name on it
Works because a human is on the hook for the recommendation, not because the channel is clever.
Short video where the demo is the ad
The product proves itself in eight seconds. This is evidence, compressed.
Being in the model's answer
Earned by being discussed where models read, which you cannot write yourself.
Working in public
People learn your name before they learn your product. Slow, and therefore hard to copy.
Word of mouth inside a profession
Compounds, has no algorithm, does not deflate.
Read the right-hand column carefully

Nothing on that list is safe as a tactic. Creator distribution and short video are already being flooded by generated ad farms and faceless channels, exactly as search was. The durable line is not between old channels and new ones, it is between attention you rent and attention you own. Rented attention gets repriced. Owned attention does not.

07 Three forks, and what each one costs
Fork one

Up the ambition

Use agents to reach the integrated scope that used to require a team. This is the direct answer to the capacity claim, and it is now genuinely available.

The trade Build time is the same time distribution now demands. Fails when A large product nobody finds, and a support load that consumes the founder. Honest headcount Two to four people. Not one.
Fork two

Audience first, product second

Invert the order. Earn attention in a narrow group, then build the thing that group has already asked you for twice. Distribution stops being a launch problem.

The trade One to two years of unpaid audience building before revenue is plausible. Fails when You build an audience that likes you but does not buy, or you become the channel and can never step away. Honest headcount One, but the one has to be visible.
Fork three

Sell outcomes, not seats

Let the agent do the work. Charge for the result and carry the liability. Distribution here is a professional network, which is cheap and compounds.

The trade It is a services business wearing software margins, and it does not resell without effort. Fails when You have bought yourself a job with worse hours. Why it holds What is being sold is accountability, which no model can assume on your behalf.

Verdict

Not dead. Repriced.

What deflated is the part that was never scarce. What did not deflate is trust, attention, and being the one who is answerable when it breaks. The category's failure rate was around ninety-nine percent before agents existed, and it was distribution that set that number. Making the build free changes the composition of the survivors, not whether there are any.

The case above is right about one thing and wrong about why. The generic middle really is being cleared out. It is being cleared by the labs owning the default surface people already have open, and by state and context migrating into the model, not by the fact that code became easy to write.

The two prescriptions also pull against each other. Build bigger, and distribution is now most of the job, compete for the same week in the same calendar. Only one of them can win, which is why the realistic shapes are a small team rather than a solo founder, or a smaller product sitting behind an audience already held.

The bar moved rather than closing. It used to be: can you build this. It is now: can you get in front of the right hundred people, and will you stand behind the result. Both of those are harder than they were. Neither is a wall.