100 Mistakes at a Meeting or "Not Passed": How to Master Speaking Faster Than "Learning a Language" (Including with AI)
You can spend years learning a language and still freeze on the first call from a client. Or you can shift to a mode where language stops being a school subject and becomes something practical: a tool for relationships, work, and identity.
The mechanism is simple: instead of chasing "B2", the number of words and perfection, I build speaking competency – so that in a real-life situation I can convey what I want to convey. And if I make 100 mistakes along the way, even better.
"Language" is a tool. "Speech" is a relationship.
The very word "language" sets up technical thinking: grammar, words, rules. And then, surprise – that approach doesn't work in a real conversation.
That's why I use the term speech – because speech immediately assumes another person. In speech, you see a face, intention, context, gesture, pace, hierarchy, and emotions. Language is people: it's their stories, traumas, pride, jokes, and ways of being.
This is especially important today, when AI translates words but doesn't translate relationships.
Competence has 3 elements: attitude, knowledge, skill
In learning "school language", attitude usually suffers most. Classic scenario: grades, tests, "you need to understand perfectly", zero room for mistakes. Effect? Demotivation.
I build competence like any other – from three elements:
- Attitude (approach, courage, goal, identity)
- Knowledge (rules, explanations, vocabulary – but useful)
- Skill (4 domains: listening, speaking, reading, writing)
If even one element is lacking, the whole thing collapses. Most often it's attitude and speaking practice that fail.
Where do "13 languages" come from? Biography, not counting
The number "13" is impressive, but in practice it's not collecting. It's the result of languages appearing in specific life roles.
An example path that shows this logic perfectly:
- Startup surrounded by Russian, Polish, and Latvian (Latvia + Polish school),
- English as a school standard and "access language" to materials,
- Japanese as five years of immersion: studies, environment, senseis – the "unlocking" experience,
- Azerbaijani as the "language of love" (married to someone from Azerbaijan); Turkish comes along too,
- Further languages "from work" (sales trainer, international teams): Estonian, Lithuanian, French,
- Then semi-work/semi-private: Italian, Ukrainian, Czech (Czech up to the level of running training sessions).
There's no magic here. The rule is: a language comes in when it has a reason to come in.
Fluency isn’t equal in all languages (and “7 at once” is often the ceiling)
You can have many languages "in your portfolio", but not all are at the same level of public speaking, training, or negotiation at the same time.
There is a practical limit: about 7 languages maintained simultaneously at a very high fluency. The rest operate in a functional mode (I can manage, read, watch), and when needed, a language can be "de-rusted" with routine – sometimes in as little as two weeks of intensive contact.
And yes, languages can get mixed up:
- Italian pulls in French words (sometimes "Italianizing" saves the day),
- Surprisingly, Estonian and Japanese can overlap (grammar vs pronunciation),
- Lithuanian and Ukrainian can also mix.
That's not a flaw. It's a result of speech being contextual. When I switch to one language, I disconnect the others and create an environment where the brain has no doubt which "world" I'm in.
Task-based immersion: I don't count words, just effectiveness at the meeting
The question "How much can I learn in a weekend?" is wrongly framed; it forces you to count. A better question is:
Will I be effective at a specific meeting and convey what I need to convey?
If I have a meeting in Czech on Tuesday, starting Friday, I go into "boot camp" mode:
- I watch and read only in Czech (podcasts, shows, books),
- I use shadowing, talk to myself, read aloud,
- I listen to hip hop/rap, which contains lots of language material: rhythm, rhymes, words,
- I talk to AI only in the target language,
- I take a walk and do live dialog with Gemini (formerly often ChatGPT), talking through my meeting plan and checking where I lack words.
The key: I prepare topics and content that will actually come up. The language is supposed to serve the task, not a test.
AI gives a boost – as long as it doesn’t steal the “aha moment”
AI can dramatically speed up speaking practice because it provides something that was missing before: an instant training partner for 30 minutes in exactly the topic I need.
There's also a trap: AI creates cognitive ease and can start doing things for you too early. The order makes a difference:
1) First, a moment of uncertainty (your brain tries to figure it out from context), 2) Then the AI answer, 3) Finally – examples, paraphrases, consolidation.
If you always just "translate for me and paste", your brain has no reason to build patterns.
How to profile AI to help beginners (not frustrate them)
This is pure prompt work and setting rules:
- Tell the AI: "I'm at A2 level, speak slowly, use simple sentences"
- Ask for feedback and summaries
- Speak "like a child", make mistakes, then have AI:
- paraphrase your statement "like an adult"
- after each of your sentences, reply: "If I understood you correctly, you said..."
- Ask for texts at A1/A2/B1 level and for difficulty jumps when you grasp it
Absurd sentences work great for consolidation because novelty is coded better (e.g. "about a family of beavers who flew to the States"). Sounds silly – really works.
"Coffee" shouldn't be "kafie". It should be a concept, not a translation.
Constant word-for-word translation does unnecessary wiring in your head: target language → native language → meaning.
A better model: I connect a new word to a concept, not a native equivalent. So, not “coffee = kafie”, but “kafie” goes on the same mental shelf as: aroma, cup, morning, energy, ritual.
Comprehensible input works great here: materials where I get the meaning from context (gestures, situations, images), without translating every word. That’s when a "new shelf" is built in your brain, not just dictionary mapping.
At the same time, as an adult I add turbo: I take one sentence that caught my interest, throw it into AI and ask:
- Explain the grammar normally,
- Explain it like I'm five,
- Give 5 usage examples.
Adults don’t have to pretend to be kids 100%. Kids learn for years.
Sapir–Whorf hypothesis in practice: language switches personas (especially Japanese)
The Sapir–Whorf hypothesis says language influences how we perceive the world. There's debate about cause and effect (language vs. history/geography/culture), but in practice, you can see one thing: switching language brings out different edges of personality.
You see it most in Japanese, because the grammar enforces relational thinking. Before I say a sentence, I have to assess:
- Is my conversation partner higher/lower in hierarchy,
- Age,
- Role (e.g. CEO),
- Distance and relationship.
The choice of verb forms “humbles me” and “lifts you up.” And then body language too: bows, different postures, different economy of “I” (in everyday Japanese, "I" doesn’t push to the center).
Side effect can be lasting: one returns from years in Japan and even in native language keeps "apologizing and thanking all the time".
In the other direction, French brings out "c’est moi" energy – more self-confident, more "on the front line".
And here the AI theme returns: automatic dubbings and translations often lose what matters most. The Italian ragazzi changed to English "boys/guys" destroys context. It's not the same relational nuance.
Perfectionism kills speech. The antidote: 100 mistakes
Perfectionism usually has psychological and school roots, but in practice, one simple switch works:
At the meeting, the aim is to make as many mistakes as possible. At least 100 – otherwise, it's "not passed".
Logic is merciless: 100 mistakes = 100 tries = 100 iterations. From this, you can extract concrete corrections. Without attempts, there’s no data – and without data, no progress.
How to measure progress without the certificate obsession
Instead of "do I have B2?", use functional tests:
- Go back to a song you listened to 3 months ago and suddenly you’ll notice patterns (verbs, constructions),
- Check how long you can keep up a conversation,
- Have some small talk riddles prepared (name, "what's up", one question) – this breaks the ice,
- Do recordings: once a month, record your speech on the same topic and compare how you sound.
Today you can do this with AI too: voice conversation as a test "can I manage", plus correction and paraphrasing.
Case: "bus polyglot" and 800 hours a year with no extra time
The biggest myth is "I don't have time." The time is usually in the gray zone of logistics.
A model that shows this perfectly:
- Bus commute: 40 minutes one way and 40 minutes back,
- Over a year that's over 800 hours of exposure.
That’s exactly the scale that makes a difference – similarly, bilingual preschools often have "absurd" numbers like 800 hours of English a year (that is: everything happens in the language).
In practice:
- I set the language on my phone,
- I shove the language into my ear on walks, in the shop, on the go,
- I accept that at the start it’s gibberish and uncomfortable,
- I keep the routine until my brain "surrenders" and starts seeking patterns,
- I add people: if there are folks from Tunisia or Belgium at work, I say: "Please speak to me in French".
A language becomes tied to faces and stories, not just to a vocabulary list. Example of relational power that opens doors in a second: talking to someone from Congo in French and asking: "Do you speak Lingala?" – even without knowing Lingala. The mere signal "I know your context" shifts the trust filter.
Takeaways that work in real life
- The goal should be about identity and situation, not numbers ("B2" is not tangible).
- Competence is built from three parts: attitude + knowledge + skill.
- Input is the foundation: songs (often rap), shows "for natives", comprehensible input.
- AI speeds you up if it provides: paraphrase, feedback, level adjustment – not just instant crutches.
- Mistakes are a KPI: without them there’s no iteration, and without iteration, no speech.
- "I don't have time" often means "I haven’t counted my gray zone" (40+40 minutes a day = 800h/year).
How to implement this (steps to start right now)
1) Define a goal as a scene, not a level - Instead of "B2": "In 12 months I’ll run a 30-minute status meeting in French and finalize arrangements."
2) Plan 2–3 fixed "gray zones" per day - commute, walk, shopping, washing up. At least 30–90 minutes of exposure daily.
3) Build three input channels - playlist of 30–50 tracks (rap/hip-hop welcome), - 1–2 podcasts/formats for natives (not for learners), - 1 comprehensible input format (gestures, situations, simple dialogues).
4) Set AI as a coach, not a translator
- Starting prompt (in the target language):
- "I'm at A2 level. Speak slowly. After each of my statements, write: ‘If I understood you correctly, you said…’ Then provide a correct paraphrase and one short tip."
- Once a day, 10 minutes of conversation about what you’ll actually do today/tomorrow.
5) Introduce the 100 mistakes rule - In every speaking session, the goal is number of attempts. After the session, write down your 5 most common mistakes and ask the AI for 10 example sentences.
6) Measure progress functionally every 2 weeks - record 60–90 seconds on the same topic, - revisit the same song and check the lyrics, - do a 5-minute small talk (with a person or AI) on the same questions.
If this system is to work, one element is non-negotiable: speech must enter real life, not stay just in the app. Then even imperfect sentences do what they’re supposed to do: connect people and deliver results.