The work is getting cheaper, and a lot of people selling it have not said that out loud yet.

A year ago you could charge a real fee to connect two apps and call it an AI system. The client had not used a capable model. You had. That gap was the business. It is closing. Models that scored nothing on automation tests are clearing half of them now, and the monthly improvement is steady enough that the rest will not take years. You can already hand a messy site to an agent and get a first pass back without anyone dragging boxes around a builder. The premium for the simple connection is what goes first.

Arbitrage was always the product. You got paid for knowing something the buyer did not. When most of the room has tried the tool, that fee shrinks, even if the tool is better. Staying on the same offer while the floor rises is not neutral. You get paid less for the same afternoon.

That does not mean the work disappears. Accounting, marketing, sales, and the back office are all being rewritten. If labor is being replaced, the people who know where it should be replaced still have a job. The job is just no longer “I will build the Zapier flow.”

What still gets a check is easier to see if you stop defending the old menu.

Clients will pay you to tell them what not to automate, and to show a team how to use the thing you would have built for them last spring. Implementation of a basic stack is a workshop now. The workshop is the offer. Saying what the offer actually is, in their words, matters more than another certification. Website builders did this to coding years ago. Plenty of people still sell expensive sites. They sell the judgment and the pitch, not the login to Wix.

Text and a flat image are already a commodity. Video, a real product visualization, a simulation a buyer can walk through, those still have a price, because they are harder to fake in a afternoon and harder for the client to prompt on their own. If your deliverable is a paragraph, assume the fee is on the way down.

Drag-and-drop builders are the same story. Zapier, Make, and Power Automate still run a lot of small shops. They are a weak thing to lead with once the client has seen an agent write the glue itself. Custom tools, internal apps, and coding agents such as Claude Code or Codex are where the paid work is moving. You can keep a builder for yourself. Do not make it the thing on the invoice.

A general “we do AI” agency is the first offer an agent can imitate. A person who knows how a clinic handles records, or how a warehouse actually loses a day, is harder to imitate. The model can draft the workflow. It does not sit in the compliance review. Pick a vertical and learn the boring constraints. That is the moat, not the prompt.

The other invoice that shows up later is the audit. Once a company has agents in the process, someone has to ask what they are allowed to see, what they stored, and who is liable when the output is wrong. That work looks dull next to a demo. Dull is what legal and security will pay for when the demo has already been bought.

Owners who feel late do not want a chatbot to talk them through it. They want a person in the room who can sequence the change so the staff do not revolt. That is change management with a plainer name. Running the company so it does not depend on one hero is the same muscle. If you can do that conversation, you will outlast the person who only sells the build.

Some of the money in the next year will be short. A new gap opens, a few people sell into it, the model catches up, the fee dies. Take those if you can see the end date. Do not build the company on one of them.

The intermediaries who only passed information from a tool to a client are the ones who get skipped. The ones who can still tell a business what to automate, what to leave alone, and how to get the team to use it, have the better few years. The tool will keep getting cheaper. The decision will not.

Heres a great video by Nick Saraev on what he would learn instead of AI automation

There is a moment in almost every cross-border meeting that nobody brings up afterward. Somebody says something that matters, the other side nods, and 3 weeks later you learn the nod meant nothing.

Your best engineer works out of Warsaw. Your supplier runs out of Shenzhen. Your new distributor is in São Paulo. The talent is global, and the revenue is global. The meeting still runs in a single language, and everyone outside that language works at half speed.

That gap costs real money, and most founders never see the invoice for it.

The Quiet Cost of a Polite Nod

Nobody stops a client call to say they did not follow the last 2 minutes. They catch the gist, fill in the rest later, and hope the gaps were small ones.

So the quietest person on the call is often the one who understood the least, and sometimes the one who knew the most about the problem. You hired that person for their judgment. You get their silence instead, because forming a sentence in a second language under time pressure costs more energy than most fluent speakers ever notice.

The commercial version of this shows up later. Grammarly’s State of Business Communication research found that 1 in 5 business leaders had lost business over communication problems, with the majority putting the value of what they lost at $10,000 or more. Those numbers cover communication generally. Add a language barrier on top, and the misreads get bigger and slower to surface.

Captions Solved the Wrong Half of the Problem

Most teams reach for translated captions first, and they do help. But they also make your brain read and listen at the same time while the conversation moves at full speed.

Try it during a screen share. The deck takes most of the screen, captions sit at the bottom, and the speaker keeps talking. Something gets missed, usually the nuance that matters.

Captions also create a bigger business problem: they disappear. Once the call ends, the translated version is gone, leaving a recording in a language half your team cannot easily review. When someone later asks what the supplier actually committed to, you are left relying on memory and interpretation.

What Changes When the Translation Is Spoken

An AI interpreter does what a human interpreter does, which is speak the translation aloud in the listener’s language while the conversation keeps moving. Your counterpart talks in Japanese. You hear English. Your reply goes back the other way. Nobody reads anything.

The difference in a live negotiation is sharper than it sounds on paper. Tone survives. Hesitation survives. When a supplier gives you a soft refusal wrapped in politeness, you hear the hedge in the delivery rather than reading a flat sentence that looks like agreement.

Tools in this category have moved quickly. JotMe, for instance, runs live translation across 200+ languages and 39,000+ language pairs, speaks the translation aloud in 100+ languages, and can carry the speaker’s own voice into the translated audio. It runs as a desktop app alongside Zoom, Microsoft Teams, Google Meet, and Webex, capturing audio on your machine rather than joining the call as a bot. 

The Built-In Options Have a Ceiling

Every major meeting app now advertises live translation, and the headline hides a short list.

The table also highlights important limitations around language coverage and licensing. Language support can fall short when your business operates beyond major European and East Asian languages. More importantly, native features are typically tied to the meeting organizer’s licence. If your client hosts the call, your own paid plan may offer little to no benefit, even when you are the one who needs the transcription or translation. For cross-border meetings, where the invite often comes from someone outside your organization, this licensing model can quickly become a significant limitation.

What to Check Before You Pay for Anything

• Does it follow you or the host? A tool running on your own device works on any call. A built-in feature works only when the organizer pays for it.
• Does it speak, or only caption? Ask for the spoken language list specifically, because it is always shorter than the caption list.
• What survives the meeting? Confirm you keep the recording, the source transcript, and the translated transcript, because most built-in captions are gone the second the call ends.
• Does anyone see a bot? On client and candidate calls, a tool that runs quietly on your machine avoids a conversation you do not want to have.
• Can your team test it for free? Run a real call in your actual language pair before anyone signs anything. 20 minutes with a live supplier tells you more than a feature page.

Final Thoughts

The global workforce is already here. Your team, suppliers, and buyers speak more languages than any one person can, and hiring around that gap means choosing fluency over ability.

AI interpretation removes that constraint. The person with the sharpest insight into your supply chain can share it in their own language, while the decision is still being made. That is the real value.

Pick one recurring cross-border call this week and turn on live translation. See who speaks up that usually stays quiet. That costs nothing and tells you whether this is worth budgeting for.

AI visibility is only valuable when the information being surfaced is accurate.

A brand may appear in ChatGPT, Gemini, Perplexity, or Claude but still be described with outdated products, incorrect positioning, or incomplete information. For companies relying on generative search as part of customer discovery, those inaccuracies can create confusion and weaken trust.

Generative engine optimization (GEO) can help brands improve both visibility and the quality of how they are represented.

Here are five companies specializing in generative search optimization for improving AI recommendation accuracy.

1. Profound

Profound is a strong fit for companies that want detailed visibility into how AI platforms describe their brands.

Its platform focuses on areas such as brand visibility, citations, sentiment, competitive performance, and answer accuracy across generative search.

That makes it useful for identifying whether AI systems are presenting outdated or inconsistent information.

Companies can monitor how descriptions change across prompts and platforms, then use those insights to determine where corrections or stronger source signals may be needed.

For organizations with internal marketing and content teams, Profound can provide a strong measurement layer around recommendation accuracy.

2. iPullRank

iPullRank is particularly useful when inaccurate AI results stem from unclear digital information.

Large websites can create conflicting signals when products, services, or categories are described differently across multiple pages.

iPullRank can help improve technical SEO, information architecture, entity relationships, and content structure.

For GEO, that can reduce ambiguity around what the company offers and which information should be considered authoritative.

This is especially relevant for enterprise brands with complicated product portfolios or years of legacy content.

3. NP Digital

NP Digital is a strong option for companies that want to investigate why AI platforms are getting their brand story wrong.

An inaccurate recommendation may be a symptom of a broader issue.

Old third-party articles may still describe a previous product offering. Different pages may use conflicting terminology. Competitors may have stronger category associations, causing AI platforms to position the brand incorrectly.

NP Digital can help identify where those inconsistencies originate and determine which parts of the digital footprint need attention.

That could involve updating high-value content, improving technical clarity, strengthening entity signals, or earning newer third-party coverage that better reflects the company’s current positioning.

For brands that have evolved significantly over time, this can make GEO an important part of reputation and information management.

4. Omnius

Omnius is a good fit for SaaS, fintech, and technology companies that want specialized generative search analysis.

Its GEO focus can help brands monitor how products and services are represented across AI-generated recommendations and comparisons.

That is particularly useful in fast-moving categories where features, pricing models, and positioning change frequently.

If AI platforms continue surfacing outdated information, Omnius can help identify which prompts are affected and where content or authority improvements may be needed.

5. Reboot Online

Reboot Online can help companies improve recommendation accuracy by strengthening current third-party information.

Brands cannot control every external source discussing them, but they can create new, credible information that better reflects their present positioning.

Digital PR can help accomplish that.

Original research, expert commentary, and relevant media coverage can expand the number of recent external sources associated with the company.

For GEO, that can help strengthen a more accurate digital picture over time.

Why AI Recommendation Accuracy Matters

An AI mention is not automatically positive.

If a platform describes a company incorrectly, recommends the wrong product, or associates the brand with an outdated category, the result may hurt rather than help.

Companies should therefore monitor both frequency and accuracy.

Important areas to review include product descriptions, target audiences, pricing, capabilities, locations, leadership information, and category positioning.

How Brands Can Improve Accuracy

The first step is identifying where inconsistencies exist.

Owned content should be updated and aligned around current terminology.

Important third-party listings and profiles should also be reviewed where possible.

Brands should make core information clear and easy to find rather than forcing search systems to infer important details. Structured data is one practical way to do this, since it labels key facts about a business in a format machines can read directly.

New external coverage can also help reinforce current positioning when older information remains online. 

Choosing the Right GEO Company

The best partner should help diagnose why inaccurate recommendations are happening.

Some companies need stronger technical structure. Others need clearer content, updated entity information, or more current third-party authority.

A strong GEO company should improve not only how often the brand appears, but also how correctly it is represented when it does.

The AI race is gradually entering a new stage, where investors are beginning to assess not only growth rates but also the ability to turn huge computing costs into sustainable profits. Against this backdrop, Anthropic looks particularly interesting, as the company claims it will end the current quarter with adjusted operating profit for the second time in a row and is simultaneously gearing up for an upcoming IPO with a potential valuation of up to $2 trillion.

In the second quarter, Anthropic already reported adjusted operating profit, and its revenue grew 14‑fold year‑on‑year to $11.5 billion. As of the end of July, its annualized revenue run rate reached $65 billion, compared to less than $9 billion at the end of the previous year. If the current trend continues, the figure is expected to climb to about $120 billion this year, with nearly threefold growth next year.

The business’s economics look particularly impressive before accounting for the most onerous infrastructure costs. According to sources, without the costs of cloud partners, model training, and a number of other items, Anthropic’s gross margin exceeds 80%. However, it is precisely these expenses that remain a key issue for future public investors. Developing advanced models requires ever‑increasing computing power, which means high capital intensity and constant access to infrastructure.

Nevertheless, the move to operating profitability already sets Anthropic apart from the typical image of an AI startup that has been burning capital for years in pursuit of growth. It is the combination of strong revenue growth and the first signs of profitability that could support a valuation of up to $2 trillion.

By contrast, OpenAI is taking a more cautious approach. Sam Altman has officially confirmed for the first time that the company will not be conducting an IPO in 2026. Formally, the reason was the tension surrounding AI safety, but for financial markets, this creates an important contrast. Anthropic aims to leverage its current strong performance to launch an IPO, while OpenAI prefers the flexibility of remaining private.

The risks here are indeed becoming increasingly noticeable. Anthropic’s management is proposing an independent audit of safety procedures, coordination among major developers, and even a voluntary slowdown in model development. OpenAI and xAI have publicly supported some of these initiatives. For businesses, this approach potentially means slower releases, additional costs for testing and compliance, and higher regulatory risk.

That is why the timing for Anthropic’s IPO looks ambiguous. On the one hand, the company may enter the market during a period of exceptional growth and broader market optimism — reflected in recent gains across U.S. equity futures (such as ES futures) on the back of the AI rally — when its financial performance allows it to justify a premium valuation. On the other hand, investors will effectively have to evaluate a business whose management simultaneously warns of the need to limit the pace of development in its own industry.

Geopolitics adds an additional layer of complexity. In the U.S., calls for stricter security measures are met with concerns about losing the technological race to China. Donald Trump has already stated that regulation should not hinder American companies from maintaining their leadership, and some lawmakers fear that excessive restrictions will simply give an advantage to Chinese developers.

At the same time, China itself is also increasing its focus on AI security, but it is emphasizing political, informational, and infrastructural risks rather than voluntarily slowing down technological progress. This reduces the likelihood that the U.S. will be able to significantly slow down the development of the industry without harming its own competitiveness.

As a result, Anthropic’s IPO may become one of the most important tests for the entire AI market. Investors will have to decide how justified the valuation of up to $2 trillion is for a business with phenomenal growth rates, but at the same time huge infrastructure needs and growing regulatory risks.

If Anthropic can maintain high profitability and growth after going public, the market will receive the first convincing proof that advanced AI models can be not only technologically viable but also financially sustainable. If, however, safety concerns and regulation begin to limit the pace of development, the current valuations may turn out to be too aggressive even for one of the fastest‑growing companies in the world.