The market became an AI experiment

The market with AI became aggressive because companies, investors, professionals, and consumers are trying to understand a shift that has already started. Artificial intelligence already writes code, creates products, accelerates marketing, changes hiring, cuts costs, creates new opportunities, and threatens entire services, but nobody knows where this stabilizes.

The core point is simple: companies are testing artificial intelligence in the real market. The test subjects are the product, the team, the customer, the employee who stayed, the employee who left, the vendor, the candidate, and the entrepreneur.

Why did the market become so aggressive?

The market was already chaotic. AI amplified that chaos because it added speed, promise, and fear at the same time.

The market became more aggressive because five pressures arrived together:

Higher cost of capital.
Mass layoffs and pressure for margins.
Artificial intelligence promising more output with fewer people.
Companies trying to look current to the market.
Professionals trying to prove they are still needed.

This changes the conversation inside companies. Before, the question was "how do we hire more people to deliver more?". Now, in many places, the question became "how many people are still needed if AI can produce, sell, support, analyze, and code?".

That question sounds objective, but it is still poorly framed. An AI agent does not replace a whole profession as a discipline. It replaces part of the execution, accelerates part of the work, and makes some tasks too cheap to keep manual.

The problem is that the market is jumping from discovery to cuts. Many companies still do not know what to do with AI, but they already act as if they do.

At the same time, the mood is contradictory. Some people want to buy a small farm, leave the noise, and look for stability. Others are excited about million-dollar funding rounds, new companies, and the chance to build something big now. Both reactions make sense. The market is scared and excited at the same time.

Is this a revolution or a bubble?

It is both. Artificial intelligence is a revolution in capability and a bubble in narrative, valuation, and easy promises.

AI is here to stay. That is a fact. It left San Francisco (US), Shanghai (China), and the labs. It is in churches, neighborhood bars, small consultancies, legacy products, marketing, legal, support, schools, and code editors.

At the same time, nobody knows exactly what to do with it yet. Everyone is testing:

Test type	What is being validated
Product with AI	Whether users pay for a better answer
Developer with agent	Whether fewer people can keep the same roadmap
Marketing with AI	Whether "uses AI" still sells innovation
Hiring with AI	Whether companies want to measure fundamentals or tool use
Operations with AI	Whether automation cuts cost without breaking quality

Some new companies will grow large because AI lowered the cost of starting. A small team can research a market, write code, create support, produce marketing, sell, measure usage, and iterate at a speed that used to require many more people. That does not guarantee a durable company, but it increases the number of attempts.

It also changes the kind of company that can be born. A startup only a few years old can reach huge revenue if it finds a real bottleneck, uses AI to deliver better, and distributes fast. At the same time, many will disappear because they were only a thin interface on top of a model, without distribution, proprietary data, trust, or a critical workflow.

Some products without AI will keep working well because they solve mature problems. Other products will lose meaning because AI removes the need for the service. And many new things will appear where there was no market before.

The ocean may be blue, but it is rough. There is opportunity, strong wind, and broken boats.

Why does everyone want to create something now?

Before, a specialist from another field almost always needed to hire a programmer to turn an idea into software. A doctor, lawyer, teacher, consultant, or niche operator explained the problem, waited for a quote, translated the domain to a technical team, and hoped the result came close to what they imagined.

Now that specialist can test alone. They open an AI tool, describe the flow, paste examples, ask for an interface, connect data, and reach a working prototype. It may not be safe. It may not scale. It may not be ready for production. But it may be enough to validate an idea.

That changes the market because the creation barrier dropped:

Before	Now
An idea depended on technical budget	An idea starts as an AI prototype
Domain knowledge was separate from implementation	Domain knowledge goes straight into prompts and feedback
Niche software took longer to appear	Niche software appears in days
The programmer was the first door	AI becomes the first door

The clearest example appears in technical and regulated fields. A professional may want to create a viewer, a document organizer, a report automation, an analysis dashboard, or an internal tool. Before, that looked like a company project. Today, it can look like a weekend experiment.

That is the good and dangerous part. The person who understands the problem can now build. But when sensitive data, healthcare, contracts, money, or important decisions enter the system, the central question comes back: who is judging security, privacy, quality, and responsibility?

Do companies know what they are doing?

Many do not. They know they need to test, cut costs, and show productivity. That is not the same as having a mature strategy.

Entrepreneurship is doing, shipping, and validating. You do not learn that well from theory alone. You learn by putting something in the world, seeing what broke, listening to the market, and adjusting. With AI, this cycle became faster, but also more dangerous.

Now many tests are happening at the same time:

Test	Who pays the cost
Cut the team and keep the roadmap	People who stayed
Put AI into the product without clarity	Users
Promise productivity before measuring quality	Internal teams
Replace judgment with automation	Product and reputation
Use vibe coding without fundamentals	Future maintenance

The market is validating hypotheses in production. Some will work. Many will create rework, technical debt, incidents, weak products, and bad decisions.

This applies to product, hiring, management, marketing, and engineering. Companies use AI to survive, cut costs, solve scarcity of good talent, and signal to the market that they are still current. Part of that is strategy. Part of it is fear.

And fear is not irrational here. Companies that are already established need to stay relevant. Companies that are not established can use AI to move faster and take market share.

Why did developers become the target?

Developers became one of the most visible targets because code is a concrete part of delivery. If AI writes code, it looks like the cost of developers went down.

But software engineering was never only about writing code. It also means understanding the domain, modeling data, choosing trade-offs, controlling risk, reviewing security, keeping systems alive, and protecting the business from pretty solutions that break later.

Just look at a map like roadmap.sh to understand the size of the field. There are paths for frontend, backend, full stack, DevOps, DevSecOps, AI Engineer, Forward Deployed Engineer, architecture, security, databases, system design, code review, Claude Code, vibe coding, and AI agents. Each path pulls concepts, fundamentals, technologies, and tools.

And there is another layer: even when you already understand fundamentals, you still need to learn how each tool works. Knowing architecture helps, but it does not automatically teach you the behavior of Claude Code, Codex, a new framework, a new cloud, a specific database, or a model with its own limits. Fundamentals help you ask better questions, notice risk earlier, and learn a tool with more judgment. They do not replace studying the tool.

The confusion starts when a company only looks at output. If five people used to ship twenty tasks and now three people with AI ship twenty tasks, finance sees a gain. What it may not see is the quality of judgment behind those tasks.

That is why pressure increases on the people who stay. The remaining developer must produce their own work, the work of the people who left, and the promise that AI multiplies everything. If they know how to use harnesses, loop engineering, agents, skills, Model Context Protocol (MCP), and automation, the expectation can become almost infinite.

That equation does not work without process. AI increases throughput, but it also increases review surface.

How did AI make hiring worse?

Tech hiring processes were already broken. With AI, they became more confusing.

Now companies need to decide:

Ban AI in the technical test to measure fundamentals?
Require AI in the test to measure tool use?
Allow anything and judge the result?
Run pair programming with AI open?
Measure final code, reasoning, review, or the ability to fix mistakes?

Each option measures a different thing. A test without AI may ignore how software will be built in the actual job. A test with AI may hide lack of fundamentals. A test that only looks at output may reward the person who copies better, not the person who understands better.

The good question is not "did this person use AI or not?". The good question is: can this person use AI without outsourcing judgment?

The same applies to code review. If the candidate generated code that passes the tests, the reviewer still needs to know whether it is safe, simple, sustainable, and coherent with the product. AI does not remove review. It changes what review must observe.

What changes with Claude Code, Codex, and agents?

Tools like Claude Code and Codex change the game because they can already do work that used to look exclusive to an experienced developer. They read code, edit files, run tests, fix failures, explain decisions, and help keep context.

In the right hands, they code above junior level. In some workflows, with good context, a good harness, and strong verifiers, they can get close to senior-level work on bounded tasks.

But there is an important difference:

Capability	AI does it well	Still needs judgment
Write code	Yes	Yes
Refactor local code	Yes	Yes
Fix tests	Yes	Yes
Design architecture	Helps a lot	Yes
Understand business risk	Partly	Yes
Decide human trade-offs	Not reliably	Yes
Take responsibility	No	Yes

The mistake is confusing production with responsibility. AI can produce. Engineering still needs to accept or reject.

This post itself is a small example. The cover looks good because of AI. The text was organized, reviewed, and corrected with AI help. I "wrote" it, in practice I directed AI to write it, in about 20 minutes while programming two projects.

But if I do not review it, AI can also leave mistakes, generic lines, and weak conclusions. The tool increases capacity. It does not remove authorship.

Who is the vibecoder?

A vibecoder is someone who can build visually functional software without mastering software engineering fundamentals. They describe, paste, adjust, ask for changes, build screens, connect APIs, and reach something that looks like a product.

This is not useless. It is a real shift. More people can turn an idea into a prototype, page, automation, landing page, dashboard, or small product.

The risk appears when the prototype becomes a critical system without technical judgment. The vibecoder may build the interface, but may not see:

A weak data model.
Bad authentication.
Missing authorization.
Data leakage.
Hidden coupling.
No tests.
Hidden operational cost.
Dependence on prompts instead of architecture.

There are also engineers doing vibe coding. The difference is that the engineer knows when to stop, inspect, test, refactor, and turn a fast build into sustainable software.

What do companies want from AI inside products?

Companies want to cut costs and increase perceived value. That is why they are laying people off, reorganizing teams, and trying to put AI inside their own products, processes, support, reports, and software.

The problem is that "add AI" became a generic answer. Sometimes, "my product uses AI" works more as marketing than as real improvement. In many cases, there is no clarity on:

Which real user pain will be solved.
Which data can or cannot enter the model.
Which error is acceptable.
Who reviews the answer.
How to measure quality.
How to explain a decision.
How to turn the automation off when it fails.

Without that, AI becomes product theater. It looks like innovation, but only adds cost, risk, and noise.

There is also a parallel industry selling shortcuts. More creators, courses, videos, and promises appear every day. Some teach useful things. Others sell Claude Code, Codex, and agents as if they were a crystal ball. The problem is not teaching tools. The problem is selling tools as a replacement for judgment.

Why are new names appearing for old roles?

The market is also creating new roles and new names for work that partly existed already. Forward Deployed Engineer (FDE), AI Engineer, Applied AI Engineer, AI Product Engineer, and AI Solutions Engineer are signs of this reorganization.

Some of these roles are truly new. Others are new names for an old mix: understand the customer, design the solution, integrate systems, write code, measure impact, and go back to adjust.

What changed is the tool at the center of the table. Now the person needs to understand product, data, prompts, models, integration, security, automation, and delivery. That explains why companies started using terms like "AI-first". Sometimes it is a real strategy. Sometimes it is positioning for investors, customers, and candidates.

The risk is confusing a new title with new maturity. Changing the job name does not fix weak fundamentals.

Why does marketing evangelize AI so much?

Marketing often evangelizes first because it knows how to sell a narrative. With Claude Code in hand, many marketers can show a huge promise to people who still do not have enough foundation to judge the risk.

That does not mean marketing is the problem. The problem is selling a powerful tool to people who still do not know what to do with it, as if the tool could solve the lack of concepts.

It is like giving a high-end phone to someone who still does not know how to send an email, attach a photo, or fill out a form. The device is excellent. The limit is still education, fundamentals, and repertoire.

With AI, the same rule applies. The tool is strong. But without fundamentals, it increases confusion at the same speed that it increases production.

What still matters for developers?

What still matters is judgment. Technical judgment is the ability to decide what to build, how to limit risk, when to accept a solution, when to reject an answer, and when to say "this is not ready yet".

The developer who only executed tickets is more exposed. The developer who understands product, architecture, data, security, operations, and automation became more important.

The new seniority is not writing everything by hand. It is knowing how to direct systems that write, test, and review part of the work. That includes:

Writing clear specs.
Creating verifiable harnesses and loops.
Setting limits for agents.
Separating prototype from system.
Measuring quality beyond speed.
Reviewing risk before celebrating productivity.
Turning repeated failures into tests, scripts, and guardrails.

The market may try to reduce the number of developers. But the more AI enters the workflow, the more valuable it becomes to know the difference between shipped software and reliable software.

Why may 2026 and 2027 hurt?

2026 and 2027 may become years of forced renewal, restructuring, and chaos. Growth hurts. Process change hurts. Cutting teams hurts. Learning a new tool while the market demands more productivity also hurts.

The hard part is that this growth does not happen in a clean environment. It happens while companies chase targets, developers try to keep jobs, founders try to survive, managers try to explain cuts, and customers expect the product to keep working.

Some people will win a lot from this shift. Some will leave the market. Some will return to fundamentals. Some will discover they knew how to execute, but did not know how to judge. That does not make AI good or bad by itself. It shows that the technology arrived before the market built maturity around it.

Are people ready for this power?

AI gave power to many people at the same time. Power to create, write, code, sell, research, automate, publish, and compete. The question is whether biology, mind, and emotions are ready for such violent productivity.

The human body did not evolve to live in an endless sprint. The mind does not process constant comparison well when people, companies, and agents seem to produce nonstop. Emotion also charges a price when everything feels urgent, possible, and late at the same time.

This point is not discussed enough. The AI conversation talks a lot about productivity, valuation, layoffs, automation, and opportunity. It talks less about anxiety, identity, fatigue, tool addiction, fear of falling behind, and the inability to turn off.

Power increased. Emotional, technical, and ethical education did not increase at the same speed.

How should you respond to this market?

There is no point denying the change. AI already changed the cost of producing software. There is also no point accepting the story that every developer became unnecessary.

The practical answer is to build an advantage where AI alone still fails:

Area	How to respond
Fundamentals	Strengthen architecture, data, security, networks, tests, and operations
Applied AI	Use agents, harnesses, skills, loops, and MCP with judgment
Product	Understand the problem, user, metric, and validation
Communication	Write specs, decisions, risks, and evidence clearly
Review	Judge quality, not only code volume

The developer who learns AI as an execution tool gains speed. The developer who learns AI as a work system gains leverage.

Summary

The market is aggressive because companies want to cut costs, increase productivity, look current, and test AI before they fully understand the shift.

Claude Code, Codex, and agents show this change clearly in software, but the pressure is bigger than software. It affects hiring, marketing, operations, product, investment, and careers.

The test subject of this cycle is the market itself: companies, people, teams, and products.