AI, Agentic Browsers and the End of Digital Autonomy

by Posted on

Contemporary work is defined by two major characteristics: performance and productivity. Artificial intelligence now fits into this logic as an optimization tool meant to amplify our cognitive capacities πŸ™‚

For this post also, you can find all files and prompts, on my GitHub account. See https://github.com/bflaven/ia_usages/tree/main/ia_semantic_layer

However, this cognitive augmentation comes with a trade-off: access to these new powers is paid for in personal data and constant cognitive validation of each step of reasoning produced by the machine.

Until now, this asymmetry remained acceptable. As a modest functional actor in late-stage capitalism, performing and productive, this price seemed reasonable to me.

Protecting my data has therefore never been an absolute priority. Certainly, I left Facebook and Twitter long ago, but by participating in the system without deep ideological adherence, I find myself in a posture of pragmatic submission rather than radical opposition.

However, I’m progressively becoming aware that this utilitarian and individualistic logic contributes to the demonetization of political and civic action in favor of instrumental rationality. This drift presents considerable risks.

This feeling of powerlessness is widely shared among my contemporaries. Some don’t even perceive that they are both the stake and instrument of these transformations. To paraphrase a biblical formula: blessed are the ignorant, for the kingdom of AI is wide open to them.

The Paradox of Impossible Action

Have you ever reached that critical point in a discussion or situation where your feeling of powerlessness becomes paralyzing, preventing you from acting?

You’ve already convinced yourself of the futility of your intervention. You docilely fall in line behind managerial considerations, that confused mixture of economic rationality (newspeak) based on ROI or KPIs.

Pragmatism and economic rationality retain their relevance, but they cannot constitute the only criteria for evaluating an action. You can also act, refuse, or accept in the name of moral principles, personal ethics, without reference to a measurable interest.

What the Hell? What Does This Have to Do with Atlas and Comet?

Discovering these browsers functioned as a revelation, or perhaps I’ve become completely paranoid. I sense that it will soon be possible to browse the web only through an agent that will guide every action, as if I were deprived, for my supposed benefit, of the power to choose. Voluntary servitude in the name of technological progress.

It’s worth remembering that something only has the power we attribute to it. This principle applies to AI and the vectors constituted by the Atlas and Comet browsers.

Altman, Musk, Pichai appear as barons of a new addictive substance (AI as a stimulant or let’s say it, a drug), forming a cartel to negotiate prices and congratulate each other on their commercial successes. Just observe their congratulations on social networks when launching their new models, like Gemini 3.

By choosing these supposedly “friendly” browsers, I’m progressively giving up my agency, autonomy, and critical thinking. Hadn’t these human qualities already been compromised by entrusting my data to Apple, Google, Microsoft? I had timidly chosen DuckDuckGo as my default search engine, while continuing to use Chrome.

With the appearance of Comet or Atlas, the stakes are double: by adopting them, you’re preparing to simultaneously eliminate Chrome and the Google search engine, often surreptitiously installed by default.

We become aware of the force of habits and uses that have shaped our user experience. I found myself destabilized by the presence of a chatbot rather than a search engine when opening Comet and Atlas.

Fundamental Questions

The emergence of these new browsers raises questions I’ve already discussed on this blog and that everyone shares:

  • Chronicle of an announced death of SEO and… Que viva el GEO (Generative Engine Optimization)
  • Personal data protection: where do we stand? (Privacy)

1. Chronicle of an Announced Death (SEO)

I’ve decided to live in a world without Google. We must acknowledge that this transition isn’t obvious.

It started modestly: Comet became my default browser. So I no longer use Google Chrome.

From there to envisioning that Google Chrome ends up in the cemetery of obsolete browsers, alongside Netscape and Internet Explorer, is just one step that I willingly take.

Beyond this announcement of no real interest, if saying is doing, then the simple act of stating this browser change constitutes in a way the death certificate of traditional search and the progressive end of the Google ecosystem as it stands. Critical observers will see in this the end of Google’s abuse of dominant position that had tied Chrome to its search engine.

We love to overthrow our idols. Every iconoclastic gesture simultaneously contains recognition and contempt. In digital as in life, we must never assume users are acquired. Most of our contemporaries are essentially calculating consumers, constantly evaluating advantages and disadvantages. They therefore prove ungrateful, fickle, sometimes vindictive.

Convinced of acting in total autonomy and freedom, I’m only docilely following the movement, hypnotized by novelty, determined not to appear outdated. Modernity above all?

Most remarkably, this renunciation of Google probably constitutes a remote-controlled, involuntary choice, motivated only by the desire for social conformity: doing AI, consuming AI from breakfast on. It’s pathetic, in a sense.

The end of traditional search proves profoundly infantilizing: having an agent facing you that answers everything without taking offense at anything, neither spelling nor formulation.

What a gold mine it must have represented for Google to dive in real-time into users’ collective psyche. But there was at least some difficulty: no pain, no gain. In the end, you more or less had to browse result pages; with Comet or Atlas, nothing left to do. What a fabulous era.

The real danger of agentic AI lies in the demonetization of information itself. That’s why users hurl insults at each other on social networks: to experience emotions, affronts, contradictions and feel alive by confronting otherness. By dialoguing with an AI Agent, you’re just pushing on already open doors.

I had also abandoned social networks, except LinkedIn, because I must “market myself” like everyone else.

Capitalism doesn’t appreciate the past, solidarities, culture. It favors novelty and speed. It’s interesting to observe how the first generation of web entrepreneurs is now contested by young AI entrepreneurs.

“We must never lose sight of the fact that capitalist production is focused on profit. And only on profit. If electronic relays are substituted for human controls, it’s not to ‘ease human suffering,’ it’s simply because this form of production reduces manufacturing costs.” Jean De BoΓ«, Belgian anarchist.

As a good cynic, I quote him because critics of capitalism don’t only formulate absurdities.

With the creation of assistants and agents, will we delegate all our questioning? No more interiority possible? All capacity to think rationally would be externalized for the greater happiness of technology companies.

Consequently, Google’s grip on my personal data has led me to reflect. One could claim this is European thinking, respectful of personal data. So I’m not ready to surrender to the first comer, the first personal data exploiter, in this case Comet.

Note: Just as the appearance of mobile phones disrupted the writing of TV series, the use of agents presents considerable narrative potential to exploit in fiction.

Doom Sells

In contemporary academic or journalistic usage, “late stage capitalism” often refers to a new mix of:

  1. the strong growth of the digital, electronics and military industries as well as their influence in society
  2. the economic concentration of corporations and banks, which control gigantic assets and market shares internationally
  3. the transition from Fordist mass production in huge assembly-line factories to Post-Fordist automated production and networks of smaller, more flexible manufacturing units supplying specialized markets
  4. increasing economic inequality of income, wealth and consumption
  5. consumerism on credit and the increasing indebtedness of the population

Late-Stage Capitalism

“For Marc Lavoie and Mario Seccareccia, in the scenario of late-stage capitalism heading toward maximum capital concentration and total financialization, accompanied by increased social conflict due to strong economic segregation between the increasingly pauperized masses and ever richer economic power, then late-stage capitalism would progressively transform into a fully authoritarian system, necessary to contain the intensification of socio-economic frustrations. One scenario for avoiding such an outcome would be an exit from late-stage capitalism through a democratic push that would impose decisive ruptures.”

2. Privacy: Who is Responsible?

Even more worrying, delegating rationality to agents could tip us into a more anxiety-inducing scenario, one where choice would no longer be possible.

Agentic Misalignment: How LLMs Could Be Insider Threats

Source: https://www.anthropic.com/research/agentic-misalignment

Liabilities and Guarantees for Privacy for Comet or Atlas (Practical)

# values

Comet {{app_name}}
Perplexity {{brand_name}}

# values
Atlas {{app_name}}
OpenAI {{brand_name}}

# values
Claude Desktop {{app_name}}
Anthropic {{brand_name}}

Essential Questions About Data Protection

As a user of {{app_name}}, what guarantees do you offer regarding privacy, personal data protection, and GDPR compliance?

1. I have not signed any commercial contract with {{app_name}} or {{brand_name}}.

2. Obviously, I have not read the "terms of service" or the general terms and conditions that bind me as a user of the {{app_name}} software solution. Can you provide me with a comprehensive summary and highlight the key points of your terms and conditions, especially regarding data usage and privacy protection?

3. There is a complete information asymmetry here: I have no idea what you are doing with my data, while you potentially know everything about me. Imagine I grant you access to my Gmail, GitHub, LinkedIn, not to mention my Facebook, TikTok, Instagram, and Twitter accounts. Using {{app_name}} gives me the distinct impression that I am being taken advantage of, like a consumer who has just signed a predatory loan agreement and will end up bankrupt in two years. What do you, as a company, have to say in response to this?

4. Despite your denials or reassuring corporate rhetoric, let's assume my data is indeed being used by your algorithm or to train your LLM. What can I do about it? Can I contact someone directlyβ€”and I emphasize directlyβ€”meaning a human being, not a chatbot or AI support at {{brand_name}}, to request the removal of my personal data? Can you provide me with an email address or phone number, as is legally required for an information site where the editor's responsibility is legally engaged?

5. What guarantees do I have that you will not use my personal queries and data to train your algorithm, your LLM, and then profile me, as Google, Amazon, Facebook, or Twitter do for surveillance and control purposes? I say this because I am a source of information that could be exploited against my will, manipulated, and influenced in my political choices, consumer habits, ideological beliefs, or even blackmailed, as intelligence agencies might do. These manipulation tactics have been theorized, for example, by British intelligence during World War II with the MICE theory (Money, Ideology, Constraint, Ego) or the FSB's Kompromat.

6. Let's consider a concrete scenario: while browsing my emailsβ€”which you now have access toβ€”you discover that I am committing or planning to commit illegal acts according to the laws of the European country I reside in. For example, preparing a terrorist attack, trading child exploitation material, or buying/selling drugs. What do you do? Will you report me to the competent authorities in my country of residence, potentially leading to an investigation and my arrest?

Let's explore a few more examples:
- I engage in political activities that you disapprove of.
- On a personal level, I exhibit moral behavior you disapprove of, such as being unfaithful to my spouse.
- I have an incurable illness that I am hiding from my family, my bank, and my insurer.
- I collect photos or drawings deemed "offensive" according to the laws of the country where the company that created you is based.
- I express views about a religion, political regime, or country that you find offensive.

By what criteria will you judge me? What are your opinions and convictions? Which jurisdiction do you fall under?

7. Just one last question, given my concerns: Do you think I should use {{app_name}}? And do you think I am objectively a good customer, or just a troublemaker waiting for an opportunity to sue you? Let me reassure you right awayβ€”I don't have the financial means to take you to court.

Two Complementary Reflections

Several contemporary postmodernist obsessions characterize our era:

The obsession with decadence and, consequently, its corollary: unwavering faith in growth.

The rejection of old age and the praise of youth worship.

These phenomena ultimately boil down to a single idea: fear of death and the dictatorship of the present.

AI perfectly illustrates these obsessions.

1. A Hint of Decadence

The same goes for the United States, China, or Russia: the fear of decadence, rejection of even a hint of weakness.

The threat to Europe’s technological sovereignty seems obvious. But after all, can’t we do without AI?

In many articles, we perceive this anxiety of an aging Europe losing steam, unable to defend itself, unable to position itself in the ongoing technological race to master AI.

“Europe observes it with paradoxical anxiety. We talk about ‘digital sovereignty’ and ‘strategic autonomy.’ What are truly irreversible cessions of sovereignty are presented as technical decisions about ‘modernization’ and ‘efficiency.'”

Source: https://elpais.com/opinion/2025-01-17/tecnopopulismo-contra-democracia-el-momento-decisivo-de-europa.html

2. The Product Owner’s Raison d’Γͺtre

I’m going through strong turbulence in my AI project and I can’t elaborate on the subject. Moreover, any account I could give would only be partial, representing only my point of view. We’re far from desirable objectivity. To achieve it, all stakeholders would need to be able to express themselves. This is precisely the raison d’Γͺtre of the Post-Mortem ritual.

As a Product Owner, you’re supposed to have the right to decide the order and relevance of developments based on complexity and business value, to challenge functionalities, propose variations, in short, to pivot with more or less flexibility. However, when the situation becomes tense, the organization takes refuge in a known and often obsolete operating mode, out of fear and cowardice.

Note: If you ask an AI to help you position yourself strategically, it can analyze the effect a message would have on the person it’s intended for. One thing is certain: not all truths are good to tell, but that doesn’t prevent you from taking back control of the narrative, if only for yourself.

Here’s the diagnosis for the project occupying me. If no intervention takes place, here’s what will happen:

# PROGNOSIS WITHOUT INTERVENTION

In 12 months:
- API maintained minimally
- Team disengaged or gone
- Project catalogued as "technical failure"
- IT says "we told you so"
- Management moves on to something else
- Nothing has changed

It’s remarkable to be so clear-sighted about a project’s future, but the same causes produce the same effects. So it’s not surprising that AI, as a strategic decision-support agent, is “clairvoyant” about this type of problem. LLMs have assimilated all the writings of strategic consulting firms. The famous Big Four (Deloitte, EY, KPMG and PwC). Besides, why buy their services? A paid license to Claude, ChatGPT, Perplexity, Mistral, etc. is enough.

Why does digital transformation fundamentally fail? Because it disrupts the baronies (management) and worries the executors (employees).

Ultimately, the bottom-up strategy is often a lure. An organization’s defense very often happens through retreat into a pyramidal functioning with a chief, in a very patriarchal logic, especially in crisis moments.

We want a chief who decides everything. We can no longer think.

Parable of the Wolf, Flock and Shepherd in the Pyrenees

  • The wolf represents innovation
  • The sheepdogs represent management
  • The sheep or flock represent employees
  • The shepherd represents management
  • The EU represents an investor or supervisory body

The wolf threatens both the flock, questions the shepherd’s management and worries the sheepdogs. We must react. The ewes lose their lambs, milk dries up due to stress (bad for cheese), panics precipitate many individuals into precipices (bad for meat).

The income from exploitation requires reacting, either by requesting more subsidies from the EU.

If war is declared against the wolf, it must be eliminated as quickly as possible to return to the status quo, preferably discreetly, at night. However, beware the return of the repressed. If someone speaks of the wolf’s possible return and questioning of management, must we shoot them too like the wolf? What to do about the growing pressure from ecologists, from society in general, sensitive to defending wildlife, while not really worrying about concrete consequences on agropastoralism whose flocks they only glimpse from afar during their summer hikes?

The Three Organizational Pathologies

1. Innovation Theater: The organization simulates innovation but systematically neutralizes it at its source.

2. Organizational Antibodies: The organization rejects change like a foreign body, developing automatic defense mechanisms.

3. Defensive Routines: Protection of incompetence through attack and inversion of responsibilities.

THE 3 ORGANIZATIONAL PATHOLOGIES - CONSULTANT GUIDE

═══════════════════════════════════════════════════════════════════

1. INNOVATION THEATER

DEFINITION
Set of practices that give the APPEARANCE of innovation without accepting its real CONSEQUENCES. The organization creates structures, budgets and pro-innovation discourse, but systematically sabotages their completion.

ORIGIN OF THE CONCEPT
Term popularized by Steve Blank (Silicon Valley entrepreneur) to describe large companies that copy innovation rituals (labs, hackathons, incubators) without accepting the philosophy (fail fast, iteration, autonomy).

CLASSIC SYMPTOMS

A. Weak signals
   - Budget allocated for "innovative project" βœ“
   - Team assembled βœ“
   - Apparent technical freedom βœ“
   BUT
   - Blocking validation from stakeholders
   - Prohibition of continuous delivery
   - No direct contact with end users
   - No success metrics measured

B. Double discourse
   General management: "We invest in innovation, strategic priority"
   Middle management: "Yes but we need to validate, control, document first"
   Result: Decision paralysis

C. Behavioral markers
   - Celebration of "implementation" rather than results
   - Focus on processes rather than outcomes
   - "We have an innovation team" becomes more important than "Our employees use the innovation"
   - PowerPoint about innovation > Real innovation

WHY DOES IT EXIST?

Psychological reasons:
   - Reduce anxiety: "We're doing something about change"
   - Ego protection: "We're modern" without questioning the existing
   - Political compromise: satisfy demand for innovation without disrupting baronies

Structural reasons:
   - Unresolved conflict between "explore" (innovation) and "exploit" (existing business)
   - Managerial KPIs incompatible with innovation (zero risk, predictability)
   - Culture of conformity > culture of experimentation

TYPICAL CORPORATE CASES

Example 1: The innovation lab
   Create an isolated "innovation lab", dedicated budget, but:
   - No mandate to change existing processes
   - Results never integrated into main products
   - Exists for LinkedIn photos, disappears after 2 years

Example 2: The sterile hackathon
   Organization of internal hackathon, good ideas emerge, then:
   - "Very interesting, we'll study it"
   - Validation committee formed
   - 18 months later: nothing has moved

Example 3: The functional ghost
   New system built, technically sound, accessible, but:
   - Gatekeepers prevent end users from accessing it
   - No impact measurement requested
   - Legacy departments sabotage integration
   = Innovation exists but is neutralized

HOW TO DETECT IT?

Diagnostic questions:
   - Do innovators have the right to fail publicly?
   - Are usage results measured and published?
   - Does innovation actually modify existing processes?
   - Are there consequences (bonuses, promotions) linked to adoption?

If 3 "no" out of 4 = Innovation theater confirmed

HOW TO GET OUT?

Guerrilla strategy (if no executive sponsor):
   - Measure usage despite prohibitions (invisible analytics)
   - Identify rebels in the organization (natural allies)
   - Create fait accomplis ("sorry, 500 users already using it")
   - Communicate success directly to executives (bypass middle management)

Frontal strategy (if executive sponsor):
   - Explicitly name the problem in executive committee
   - Demand measurable OKRs on actual adoption
   - Remove blocking validations
   - Link management bonuses to user adoption

═══════════════════════════════════════════════════════════════════

2. ORGANIZATIONAL ANTIBODIES

DEFINITION
Unconscious defense mechanisms by which an organization automatically rejects any initiative that threatens its established equilibrium, even if this initiative is objectively beneficial. Like an immune system attacking a graft.

ORIGIN OF THE CONCEPT
Popularized by Clayton Christensen (Harvard) in "The Innovator's Dilemma". Organizations develop "immune routines" that protect the status quo.

═══════════════════════════════════════════════════════════════════

3. DEFENSIVE ROUTINES

DEFINITION
Set of automatic and socially accepted behaviors that allow individuals and groups to protect their incompetence, avoid embarrassment, and maintain their position, while claiming to act in the collective interest.

ORIGIN OF THE CONCEPT
Chris Argyris (Harvard, MIT), pioneer of organizational learning. In "Overcoming Organizational Defenses" (1990), he describes how organizations prevent their own learning.

PROGNOSIS WITHOUT INTERVENTION

In 12 months:
- Solution maintained minimally
- Team disengaged or gone
- Project catalogued as "technical failure"
- Blockers say "we told you so"
- Management moves on to something else
- Nothing has changed

POSSIBLE TREATMENT

Only 2 viable options:

OPTION A: Revolution (executive sponsor)
CEO/General Manager imposes brutal transformation, breaks routines, sanctions resistors

OPTION B: Exodus (leave)
Join healthy organization where innovation is not theatrical

There is no option C (change system from inside without power).

The real question is not "how to save the project"
but "does this organization deserve to be saved?"

Technical Section

Gathering a Gold Dataset

Testing the inferences for a combination model and prompt means making comparisons and taking rational or irrational decisions.

A simple aspect: automation is often discussed and AI enables automation. All the web tools we now use seem equipped with an API. This is the case with MLflow.

Launching an AI on documentation to have it do developments so that the written script truly takes advances into account: this is one of the capabilities I’ve greatly abused lately on Perplexity.

What is a Semantic Layer?

Source: https://www.ibm.com/think/topics/semantic-layer

The Purpose for Airbnb: Today, we will explore how Airbnb builds and serves its semantic layer internally and what we can learn from it. More correctly, Airbnb did not only build a layer that “simplifies interactions between complex data storage systems and business users.” They created a complete platform.

Resources

More on dbt

An Attempt with DBT for Tags Normalization

As a data engineer and developer, how to build a prototype for a semantic layer with dbt? Can you provide steps and code?

1. You will leverage locally on DuckDB database (https://duckdb.org/) and dbt-cli (https://github.com/dbt-labs)

2. For the use case, can you imagine applying semantic layer for a use case based on "STRATEGY FOR STANDARDIZING TAXONOMIES" described below:

CONTEXT AND CONSTRAINTS
- Technical environment: CMS + API (access to taxonomies by brand and language).
- 20 languages supported (FR, EN, AR, CN, RU, etc.), each language corresponding to a distinct style guide.
- Main taxonomies: thematicTags and superTags, with combinations of the type BRAND_LANGUAGE.
- Identified problems: duplicates, inconsistencies, spelling errors, lack of semantic consistency between languages.
- Major constraint: existing content is linked to current tags, requiring rigorous management of redirects and the risk of 404 errors.

STRATEGIC OBJECTIVES
1. Clean up the taxonomies by eliminating duplicates and errors, and standardizing tags across languages.
2. Optimize SEO and visibility on Google Discover through improved management of named entities (PER, ORG, GPE, EVENT, etc.).
3. Automate processes via a Python/FastAPI API with dedicated endpoints.
4. Use French (FR) as the pivot language to define reference "master tags," with other languages acting as "slave" languages.

Source: https://www.perplexity.ai/search/as-data-enigneer-and-developer-XEumTIu_Q8yXtHHMwkHGiw

What is a Semantic Layer, and Why is it Important?

A semantic layer translates raw data into consistent, reusable metrics and dimensions, simplifying data analysis. It is a tool for maintaining uniformity across teams and tools.

Benefits of a Semantic Layer

  • Consistent data definitions: Ensures that business metrics like revenue, churn rate, YoY growth, average order value, etc., are uniformly defined. This eliminates confusion caused by different teams using inconsistent definitions and helps maintain trust in the data across the organization.
  • Enhanced collaboration: Facilitates better communication and alignment between technical and non-technical teams. By providing a unified data layer, technical teams can focus on data accuracy while business teams leverage clear and accessible metrics for strategic decisions.
  • Accelerated time-to-insight: Reduces ambiguity, enabling faster decision-making. With a semantic layer, stakeholders can quickly access and analyze reliable metrics, minimizing the time spent reconciling data or resolving discrepancies.

Sample Resources

Source: https://www.datacamp.com/tutorial/semantic-layer-with-dbt

Command Line Reference

# go to path
cd /Users/brunoflaven/Documents/01_work/blog_articles/_ia_semantic_layer/semantic_layer_with_dbt/

# create project
dbt init datacamp_project

# go into 
cd /Users/brunoflaven/Documents/01_work/blog_articles/_ia_semantic_layer/semantic_layer_with_dbt/datacamp_project/

# create this file
mkdir -p ~/.dbt
subl ~/.dbt/profiles.yml

# create file into models/
touch models/base_bank_failures.sql

# create file into models/
touch models/clean_bank_failures.sql

# create two other files
cd /Users/brunoflaven/Documents/01_work/blog_articles/_ia_semantic_layer/semantic_layer_with_dbt/datacamp_project/models
touch metrics.yml
touch exposures.yml

# commands
dbt test
dbt deps
dbt build

# Build only the raw table model
dbt run --select bank_app.base_bank_failures

# Run tests (if defined) on the raw table model
dbt test --select bank_app.base_bank_failures

# Build only the aggregated/cleaned table model
dbt run --select bank_app.clean_bank_failures

# Test only the State column in the cleaned table
dbt test --select bank_app.clean_bank_failures.State

Enabling exposures in dbt: instead of Tableau, create a Streamlit app that displays elements from the dbt project.

dbt docs generate
dbt docs serve
# check http://localhost:8080

pip install streamlit 
pip install duckdb 
pip install pandas 

xcode-select --install
pip install watchdog

Additional Resources

Semantic Layer

Other

Published by Bruno Flaven

By now, Bruno Flaven works as AI Coordinator. He has been a project manager in a wide variety of Internet business applications both in Mobile and in Desktop for 20 years now. You can find more information about his professional life on his personal website (www.flaven.fr) or his linkedin profile: https://www.linkedin.com/in/brunoflaven. He is currently working In France Media Monde (FMM) mostly on mobile applications (iOS and Android). He has also worked as P.O for a Backoffice project made with Symfony. He also made few trainings to facilitate the handling of the tools that he helps to make.

Related Posts