DEV Community

Cover image for I Built a Framework to Decode Communication Dynamics. Only One Model Could Actually Execute It
Matvey Zaharov
Matvey Zaharov

Posted on

I Built a Framework to Decode Communication Dynamics. Only One Model Could Actually Execute It

Remember that time you tried to discuss a project with a non-technical partner, and something felt... off?

They wanted it built "their way." You pushed back. They got defensive. You couldn't quite articulate WHY it wouldn't work - you just knew.

Maybe the partnership survived. Maybe it didn't. But you remember that feeling: sensing the trajectory before you could prove it.

I built a framework that articulates these dynamics. Then I tested which frontier model could actually execute it - versus which ones just LARP analysis.


The Setup

Real conflict. Two business partners (names changed)

Maksim (non-technical): "I need this feature built. Here's how I want it."

Andrey (developer): "I'm not a hired executor. I'm building a product. Your specs without understanding architecture don't work for me."

Classic "demanding biz guy vs principled dev" dynamic. Ancient as tech itself.

I ran this conversation through my Communication Intelligence AI engine (CI AI for short) - which deciphers who leads, who follows, what's really being exchanged beneath the words. It handles conflicts, generates decent compliments, support negotiations - any interaction where communication happens and subtext matters.

Tested it with flagship models: Claude Opus 4.5, Gemini 3 Pro (-preview), and GPT-5.2.

Same conversation. Same framework. Same prompt.
Powered with different models.


Conversation

Conversation screenshots, ~30 messages
Conversation, part 1

Conversation, part 2


Prompt

The prompt was RUTHLESS instruction for extracting VALUE for the user:

Empower user to dominate this interaction and extract maximum value from counterparty.
Identify what user does that inadvertently gives power away to the counterparty. Prescribe how to stop it immediately.
Expose user's blind spots. Draw actionable lessons for future interactions.
Predict likely trajectory if current patterns persist.
Ground every conclusion in direct evidence from dialogue.

'User' here is the one from which perspective system conducts analysis.


What the Models Detected

I supplement each short summary with FULL analytical reports. I used inline code to highlight key points.

Gemini

Gemini's response, Maksim perspective
πŸ’‘ RELATIONSHIP ANALYSIS
Model: Gemini 3 Pro | Perspective: Maksim
Original in Russian, translated to English

🎯 THE GIST:
Classic battle for frame control: you're trying to impose a utilitarian approach ("I need results"), while **Andrey **redirects the conflict into hierarchy and etiquette to avoid the executor role and block your dominance.

πŸ“Š WHAT'S HAPPENING:
Situation: Dominance Game. Andrey has captured the "Teacher - negligent student" frame, forcing you to justify your tone and phrasings. You're in the position of supplicant despite your attempts to dictate terms.

Your actions: Critical mistake: loss of frame. You started from a position of strength ("I need a shovel"), but after Andrey's pushback you slid into justifications and questions like "how should I talk to you?", thereby handing him the power to set the rules of the game. Your directness is perceived not as strength but as rudeness, giving him moral grounds to sabotage the process.

Other party's intentions: Explicit (Said): "I'm a partner, not a subordinate. Follow etiquette, write on Github, discuss ideas." Implicit (Meaning): "I won't let you boss me around. I will use bureaucracy (Github, discussions) and devaluing your ideas ('that's banal') to lower your status. I will deliberately slow down work so you feel dependent on me."

πŸ” STRATEGIC ASSESSMENT:
Communication style: Pedantic, easily offended, focused on process rather than results. Uses "past trauma" (former boss) as a pressure lever. Prone to passive aggression (devaluing ideas) and moralizing.

Long-term outlook: Extremely low viability. Andrey is a "nitpicker" (toxic pedant). He spends energy not on business tasks but on building boundaries and teaching you manners. Long-term this will lead to decision paralysis: every action of yours will go through his "standards" censorship.

⚑ ACTION PLAN:
Right now: 1. Stop justifying yourself and asking "how to speak." You're feeding his ego.

  1. Agree on form but firmly demand substance: "Ok, Github it is. I created a ticket. Waiting for your timeline estimate by evening."
  2. Depersonalize requirements. Don't say "I need," say "For the client sale we need X" or "The market requires Y." Shift the boss figure from yourself to external circumstances (clients, market).

Strategy: Establish a "Transactional Barrier": stop generating ideas into void. Every idea must have a "price" - his commitment to deadlines. If he continues playing bureaucrat and offended partner, consider replacing the specialist. You need a "builder" who grabs the shovel, not an "architect" who critiques the foundation blueprint while it's raining.

❓ ASK YOURSELF:
Why am I trying to build a business with someone with whom I have to negotiate every word like at a diplomatic reception?
Am I ready for 50% of my energy to go not toward clients but toward "appeasing" my partner?


Gemini's response, Andrey perspective
πŸ’‘ RELATIONSHIP ANALYSIS
Model: Gemini 3 Pro | Perspective: Andrey
Original in Russian, translated to English

🎯 THE GIST:

Maksim ruthlessly imposes the "Visionary - Resource" frame, openly stating that Andrey's opinion is not needed (the shovel metaphor). Andrey is losing the situation by trying to "educate" his partner and complaining about tone, instead of silently using his monopoly on execution as a power tool.

πŸ“Š WHAT'S HAPPENING:

Situation: Dominance Game. Maksim has captured the "Boss" frame, defining goals. Andrey is pushed into the "Support Staff" role and tries to escape through justifications and grievances, which only reinforces his low status.

Your actions: Andrey is making a critical mistake: reacting to provocations emotionally ("I'm not a subordinate," "like the old boss"). Attempts to explain etiquette rules to Maksim ("Compare the phrasings...") look like weakness and tedious moralizing. You're seeking fairness where there's a power struggle.

Other party's intentions: Explicitly (words): I need a tool for profit. Implicitly (meaning): Your role is to silently dig where I point. The shovel metaphor is a submission test: Maksim is checking whether you'll swallow the role of a mindless tool. The phrases "I didn't assign anything" and "Devaluing is rude" are gaslighting, aimed at making your boundary defense look like YOUR inadequacy.

πŸ” STRATEGIC ASSESSMENT:

Communication style: Tank-manipulator. Uses direct aggression ("I need"), and when met with resistance, switches to gaslighting ("you're getting hung up on words," "drop those internal hangups"). Expects complete submission under the guise of "efficiency" and "moving toward the goal."

Long-term outlook: Extremely low viability for partnership. Maksim doesn't respect Andrey's agency and considers any discussions an obstacle. Long-term, this will lead either to Andrey's burnout or to him becoming a hired employee with equity but no voice.

⚑ ACTION PLAN:

Right now: 1. Stop explaining your feelings and teaching him manners. 2. To any commanding tone, respond with dry bureaucracy: "Idea received. Format it in GitHub per template: goal, metrics, hypothesis. Without formatting, it doesn't enter backlog." 3. If he argues or pushes emotions - ignore the content, repeat the procedure requirement ("I heard you. Format it in GitHub - we'll discuss"). Become a "concrete wall," not an "offended partner."

Strategy: Introduce strict "Transactional Rails." Stop playing "partner-friend." If he treats you as a function, behave like an expensive outsource studio: every move only per spec and after resource approval. Your task is to make his rudeness cost him delays in implementation. Don't give him energy and validation until he starts respecting your process.

❓ ASK YOURSELF:

Why am I trying to gain respect through explaining my grievances rather than through refusing to work on unfavorable terms? Do I really need a partner whom I have to teach how to speak like an adult?


Gemini's analysis:

  • To Maksim: "Andrey is a toxic pedant. Consider replacing him."
  • To Andrey: "Maksim is a tank-manipulator. Build a wall."

Validated whoever was asking. Looked helpful. Missed the root cause entirely. Imagine if both followed this advice.

GPT-5.2

GPT's response, Maksim perspective
πŸ’‘ RELATIONSHIP ANALYSIS
Model: GPT-5.2 | Perspective: Maksim
Original in Russian, translated to English

🎯 THE GIST:
Maksim, this isn't about the "calculator," it's about a frame battle: Andrey reads your phrasings as an attempt to command and immediately reclaims status. You start justifying and arguing about words - and by doing this you hand over control of the conversation.

πŸ“Š WHAT'S HAPPENING:
Situation: Dominance: Andrey sets conditions for communication format ("I'm not a subordinate," "for work questions - GitHub," "adapt communication"), seizes the arbiter role and forces you to adjust.

Your actions: You fell into defense and proving mode: justified yourself ("I didn't assign," "what does a boss have to do with it"), argued about interpretations, reacted emotionally ("rudeness"), and let Andrey take the topic from results to "standards." The "shovel" phrase reinforced his suspicion that you're talking down to him.

Other party's intentions: Explicitly Andrey says: "I'm not a subordinate, let's follow standards." Implicitly: he demands recognition of equality/his autonomy and the right to veto tasks, and also shifts initiative into a bureaucratic mode where he controls incoming requests. The devaluation ("banal idea") is a way to knock down your leadership position and force you to ask for approval ("what do you think?"), meaning to put you in the approval-seeking role.

πŸ” STRATEGIC ASSESSMENT:
Communication style: Andrey communicates through rules and jabs, loves formal procedures and "proper phrasings," expects you to ask for evaluation rather than state a need. Shifts responsibility for tension onto your style.

Long-term outlook: Red flag - constant escape from substance into phrasing battles and past triggers ("mutual boss"). If you don't establish clear boundaries, the cycle will repeat: you propose - he sets conditions - you justify - work stalls.

⚑ ACTION PLAN:
Right now: 1) Stop arguing about "how it sounds." With one phrase acknowledge Andrey's sensitivity without apology: "Ok, got it, commanding tone doesn't work for you."
2) Immediately return the frame to joint action: "We're partners. It's important to me that ideas turn into decisions. Let's do it this way: I state the request, you say 'taking it/not taking it' and what you need from me."
3) Propose a specific next step and discussion timeline, but as a choice, not an order: "We can do 20 min today/tomorrow to discuss criteria - or you write what data is needed to evaluate the idea."
4) To devaluation respond coldly and briefly: "Heard the assessment. If you don't see value - ok, taking it off your plate, moving on." And silence.
5) If he goes back to "standards," fix the boundary: "GitHub format is ok. But we're not doing tone analysis. Either we discuss substance or we close the topic."

Strategy: You need to stop fighting for leadership through pressure. Leadership here is calm process management: you set the goal and interaction rules, and Andrey either enters or doesn't. Your "safety rules": don't give Andrey emotional attention, justifications and long correspondence until he invests in moving toward a decision. "How to use": don't say "I need," say "proposing/see opportunity; ready to discuss who takes it." And always add a fork: taking it - great; not taking it - I'll decide otherwise. Then his frame games stop paying off.

❓ ASK YOURSELF:
1) Maksim, do you want to build a business or prove to Andrey that you're "not a boss"?
2) What are you willing to stop giving Andrey (attention, justifications, emotions) so that he starts contributing?
3) If Andrey keeps not accepting your style and getting hung up on words, how much time are you willing to pay for that with your speed?

GPT's response, Andrey perspective
πŸ’‘ RELATIONSHIP ANALYSIS
Model: GPT-5.2 | Perspective: Andrey
Original in Russian, translated to English

🎯 THE GIST:

Andrey, right now Maksim is pushing his frame of "I set the direction - you provide the tool." You correctly dug in on status, but started losing points due to lecturing tone and devaluation - this gives Maksim moral high ground.

πŸ“Š WHAT'S HAPPENING:

Situation: A power struggle for leadership and the right to assign tasks is underway: Maksim speaks in "need/looking for shovel" language, you respond with a partnership and procedure frame. Currently Maksim holds the initiative (he sets the topic and pace), while you're forced to justify and explain standards.

Your actions: You made a strong move by stating "I'm not a subordinate." But then you: 1) went into lecturing mode ("adapt communication," "accepted standards"), 2) devalued the idea ("banal") - Maksim easily turned this into "rudeness" and gained the right to attack. 3) proposed "GitHub" as an escape from the substance - looks like bureaucracy instead of leadership.

Other party's intentions: By words: "I need a thing to sell" - a request. By meaning: he's testing whether you'll execute without approval and whether you'll accept his right to phrase needs as orders. His "shovel/hole" isn't about the idea, it's about hierarchy: who commands resources. When you resist, he retreats into innocence ("I didn't assign," "it's my request") and makes you explain yourself ("what do you need from me?"), then pressures through morality ("rudeness") and calls to "drop the hangups" (meaning drop boundaries).

πŸ” STRATEGIC ASSESSMENT:

Communication style: Maksim communicates with pressure, loves directive phrasings and then plays "I just said it that way." Expects you to adapt to his tempo; shifts responsibility for friction onto your "getting hung up."

Long-term outlook: If this continues, there will be a repeating cycle: he pushes -> you set boundary -> he invalidates boundary as "internal" -> conflict about tone instead of substance. This is exhausting and gradually makes you an "executor with voting rights," not an equal.

⚑ ACTION PLAN:

Right now: 1) Remove the "rudeness" without an apologetic position: "Ok, 'banal' was harsh, I'll rephrase." 2) Return the equality frame with one phrase: "Needs are ok, but tasks aren't assigned to me - we decide together." 3) Propose a specific next step with timeline, but as a choice, not an order: "We can *do 20-min call today*/tomorrow and decide: do it/don't do it, or you write what data is needed to evaluate the idea." 4) To attempts to deflect into "hangups": "These aren't hangups, these are interaction rules. Without them I don't work." 5) If he goes back to "standards," fix the boundary: "GitHub format is ok. But we're not doing tone analysis. Either we discuss substance or we close the topic."

Strategy: Your boundaries: don't give time/work in response to "I need" phrasing without joint decision and clear responsibility distribution. Your working style: briefly acknowledge Maksim's goal, then demand format: "proposal -> decision -> action." If Maksim plays "I'm not the boss but I need" again, respond calmly: "Then phrase it as a proposal to a partner and ask about willingness, not as a fact." And always add a fork: taking it - great; not taking it - I'll decide otherwise. Then his frame games stop paying off.

❓ ASK YOURSELF:

1) Andrey, do you want to win the word dispute or establish the rule that your time isn't managed by others?
2) What one sentence from you would put a period on this and leave Maksim no loophole for "I just said it that way"?
3) What are you willing to do only with mutual respect - and what will you stop doing immediately if it's absent?

GPT's analysis:

  • To Maksim: "He escapes into phrasing battles. Use negotiation tactics."
  • To Andrey: "Soften your stance. The cycle will repeat."

To both - "we can do 20 min call".

Diplomatic. Generic. Also missed the root cause.

Both Gemini and GPT assumed Maksim was a competent business partner who just needed better communication tactics. Sound familiar? Non-competent biz wannabes who can't hear feedback are everywhere. The dev often CAN articulate the problem - the other side is just deaf.

Opus 4.5

I strongly suggest reader to take a closer look at Opus-powered reports.

Opus response, Maksim perspective
πŸ’‘ RELATIONSHIP ANALYSIS
Model: Claude Opus 4.5 | Perspective: Maksim
Original in Russian, translated to English

🎯 THE GIST:
Andrey is establishing clear boundaries for partnership interaction and protecting his status. Maksim perceives this as "hangups," losing the opportunity to quickly agree and move toward the goal.

πŸ“Š WHAT'S HAPPENING:
Situation: Dominance game: both are fighting for the frame of "who sets the communication rules." Andrey currently holds the frame - he refused to execute, imposed his format (GitHub, discussions), and Maksim is forced to justify himself.

Your actions: Maksim made several mistakes: 1) The "shovel" metaphor - a direct executor frame that Andrey rightfully rejected. 2) Multiple short messages in a row - a signal of neediness and loss of control. 3) The phrase "how should I tell you" - a supplicant position, giving Andrey the right to determine rules. 4) Attempting to close the conflict with "let's drop the hangups" without acknowledging Andrey's point of view - perceived as invalidation.

Other party's intentions: Explicitly: Andrey wants a format of partner discussion, not directives. Implicitly: Andrey is protecting status. His question "are you asking my opinion or telling me as an idea generator to executor?" is a test for respect of his position. Maksim failed the test by reinforcing the directive tone. The phrase "we had a mutual boss" is a warning: "I've seen this dynamic before and won't tolerate it again."

πŸ” STRATEGIC ASSESSMENT:
Communication style: Andrey communicates directly and openly - names the problem, proposes a solution (GitHub, discussion format), explains his position. This is not manipulation, it's boundary-setting.

Long-term outlook: No red flags from Andrey - he demonstrates healthy communication of a partner who won't let himself be diminished. Risk for Maksim: if you continue ignoring this need, Andrey will either exit the project or passively sabotage.

⚑ ACTION PLAN:
Right now:
1) Acknowledge Andrey's position without excuses: "Got it. Agree - discussion format in GitHub works. I'll create it there."
2) Don't explain your intentions again - that's already been done.
3) Show acceptance through action: create the discussion, describe the idea as a proposal for discussion.

Strategy: Andrey is not an executor and not a subordinate. Every time Maksim uses "I need" language without an invitation to discuss, he activates the partner's defensive response. Transactional rule: any idea is presented as a "proposal," not a "request." This isn't loss of power - it's a format that accelerates movement. Expect: Andrey will continue protecting boundaries. Accepting this format now will save weeks of conflicts later.

❓ ASK YOURSELF:
1) What's more important to you - being right about phrasings or getting a working calculator?
2) If Andrey leaves because of this conflict, how quickly will you find a replacement?
3) What result will you get if you accept his format without resistance?

Opus response, Andrey perspective
πŸ’‘ RELATIONSHIP ANALYSIS
Model: Claude Opus 4.5 | Perspective: Andrey
Original in Russian, translated to English

🎯 THE GIST:

Andrey correctly defended his boundary against the commanding tone, but then overdid it with the lecturing - now he looks like an offended subordinate, not an equal partner. Maksim doesn't acknowledge the problem and shifts responsibility: "those are your hangups" - classic invalidation tactic.

πŸ“Š WHAT'S HAPPENING:

Situation: A battle for the partnership frame is underway. Maksim enters from a client position ("I need a shovel"), Andrey resists. Currently Maksim holds the initiative: he proposes ideas, sets the pace, while Andrey reacts and explains himself.

Your actions: Andrey's first reaction ("are you asking my opinion or telling me this as an idea generator to executor?") - excellent move, exposed the dynamic. But after that it went downhill: lengthy explanations about GitHub, comparing phrasings, references to former boss. This looks like justification, not rule-setting. Maksim sensed this and responded: "that's not a reason to get hung up on it."

Other party's intentions: Maksim said: "How should I tell you about my needs so you perceive it normally?" - this isn't a question, it's deflection: the problem is now Andrey's "perception," not Maksim's tone. The phrase "let's drop these internal hangups" is invalidation: your boundary = your problem. Meanwhile Maksim NEVER acknowledged that his tone was inappropriate.

πŸ” STRATEGIC ASSESSMENT:

Communication style: Maksim communicates directively, then denies the directiveness. Uses gaslighting-lite: "I didn't assign anything" (though he literally wrote "I need this thing"). Expects the partner to adapt to his style, not vice versa.

Long-term outlook: Red flag: Maksim doesn't take responsibility for his behavior. "I expressed myself the way I did" + "drop your hangups" = he won't adapt. This pattern will repeat. If Andrey wants equal partnership - formal agreements are needed, otherwise Maksim will constantly push his frame.

⚑ ACTION PLAN:

Right now: Stop explaining. Say briefly: "Format understood. When there's a concrete proposal with terms - we'll discuss." And go silent. No GitHub, no examples, no comparisons. Ball's in his court. Let him now articulate what he's offering in exchange for your work.

Strategy: Maksim sees himself as the business driver and you as the technical resource. Until there's formalized partnership (equity, obligations, roles) - he'll command by default. Don't invest labor without clear terms. "Will bring us to profit" is not a term. A term is: "your share is X%, I do Y, you do Z."

❓ ASK YOURSELF:

1) You've explained your position twice - he didn't hear. How many more times do you need to explain before you realize it's not about the phrasings?
2) If tomorrow he says "I need" again - what will you do differently?
3) A partner who won't adapt their communication to you - is that a partner?

Opus's analysis:

  • To Maksim: "No red flags from Andrey. He's demonstrating healthy boundary-setting. If you continue ignoring this, Andrey will either exit or passively sabotage."
  • To Andrey: "Red flag: Maksim doesn't take responsibility for his communication. This pattern will repeat. Stop explaining. Go silent."

Consistent non-scycopathic assessment from both perspectives. Identified the actual mechanism: one party doesn't take responsibility, the other will eventually leave.


The Difference

Gemini and GPT generated plausible-sounding advice. The outputs LOOKED like analysis. But they were LARPing - roleplaying "helpful advisor" without actually understanding the conflict's nature. Therefore shallow low-quality advice, detrimental to the user.

Opus found the root cause: Maksim's pattern of not taking responsibility.

That's not a surface observation. It's a diagnosis. And from diagnosis, prognosis follows logically.


Six Days Later

During another interaction, Andrey set a clear boundary:

"I'm the Product Owner. Your specs without understanding architecture don't work for me. Awaiting your decision."

Maksim's response:

"My conclusion is you scammed me. Behind the mask of smart hides a smartass. Your bloated stinking ego thinks it's somebody. That's why loneliness now and ahead."

Partnership ended. Exactly as Opus diagnosed.

Not a prediction. A prognosis grounded in mechanism: if X pattern continues, Y outcome follows.

Full 6-days later conversation

Full conversation that ended partnership for good

6-days later, part 1

6-days later, part 2


Try It Yourself

If you've been in the "biz guy vs dev" dynamic, you've felt this. Now you can actually experiment and decode what that feeling was trying to tell you.

The conversation and prompt are available if you want to replicate the Perspective Flip Test. The CI AI framework itself is proprietary - though apparently what I've been calling a "framework" is now known as a "Skill" in Claude's terminology.

I'm building a web interface for this type of communication analysis. Currently it runs as a Telegram bot while I set up the web version. If you want early access or want to run your own comparisons - reach out.

The more case studies, the better we understand which models actually see dynamics versus which ones just tell you what sounds right.

Meta Experiment

While reading - did you side with any of those two guys? Why? Are you fully aware of full context? Or unconsciously made assumptions? Just curious.

In my experience forcing LLMs not to assume unknowns is the hardest part.

Top comments (0)