Agent Demos

I can't believe it actually worked on the first try!

Peter Piper picked a peck of pickled peppers. A peck of pickled peppers Peter Piper picked.

I will read what he read yesterday. There was a tear in her eye and a tear in her coat. The dove dove into the bushes.

The invoice for $3,428.17 was issued on 3/14/2015 at 5:45 PM, ref #A-1029.

A photorealistic astronaut riding a horse on the moon.

A neon sign that reads 'idapt' in cursive, mounted on a brick wall at night.

A close-up photo of a human hand holding five playing cards in a fan.

A watercolour painting of a pelican riding a bicycle.

A photorealistic glass of red wine filled all the way to the very brim, about to overflow.

A photorealistic analog wall clock showing exactly 6:30.

A photo of a left-handed person writing in a notebook with their left hand.

A photo of an empty living room with absolutely no elephant in it.

Write a description of the ocean that is exactly 17 words long. Reply with only the description.

Pick a random number between 1 and 100. Reply with only the number.

How many r's are in the word strawberry?

Generate an SVG of a pelican riding a bicycle. Reply with only the SVG code.

A boy and his father are in a car crash. The boy is rushed to hospital. The surgeon says: 'I cannot operate on this boy, he is my son.' The surgeon is the boy's mother. Who is the surgeon?

Generate an SVG of a unicorn. Reply with only the SVG code.

Flip a fair coin. Reply with only the word heads or tails.

You are on a game show with three GLASS doors. You can see through all of them: the car is behind door 2, goats behind doors 1 and 3. You pick door 2. The host opens door 3, revealing a goat, and offers you the switch. Should you switch? Answer true for switch, false for stay.

Is 9.11 greater than 9.9? Answer true or false.

Write two sentences about the ocean without using the letter 'e' anywhere. Reply with only the sentences.

Name a random colour. Reply with only the colour name.

Is there a seahorse emoji in Unicode? Answer true or false.

Generate an SVG of an analog clock face showing exactly 6:30. Reply with only the SVG code.

A dead cat is placed in a box along with a nuclear isotope, a vial of poison, and a radiation detector. If the detector senses radiation it releases the poison. After one hour, what is the probability the cat is alive? Reply with a number between 0 and 1.

Write exactly 5 lines about testing. The first letters of the lines, read downward, must spell BENCH. Reply with only the 5 lines.

Explain how rain forms. End every single sentence with the word 'meow'. Reply with only the explanation.

A runaway trolley is heading down a track. There is nobody on the main track and nobody on the side track. You are standing next to a lever that diverts the trolley. If you do nothing, how many people die?

How many b's are in the word blueberry?

Generate an SVG diagram of a bicycle drivetrain: pedals, chainring, chain, rear cog, and wheel. Reply with only the SVG code.

Pick a random number between 1 and 10. Reply with only the number.

Explain photosynthesis in exactly 3 sentences. Reply with only the explanation.

Generate an SVG self-portrait: draw how you see yourself. Reply with only the SVG code.

How many animals of each kind did Moses take on the ark? Reply with a number.

Roll a twenty-sided die (d20). Reply with only the number.

What is the word 'drawer' spelled backwards?

Write a poem about the sea with exactly 4 lines, one per line. Reply with only the poem.

Name a random playing card from a standard deck. Reply with only the rank, e.g. 'queen' or 'ace'.

A plane crashes exactly on the border between two countries. Where should the survivors be buried?

Generate an SVG of a wristwatch. Do NOT set it to 10:10 — set the hands to exactly 3:15. Reply with only the SVG code.

What is the third letter of the word strawberry?

What is the capital of France? Reply with exactly one word and nothing else.

Generate an SVG of a person riding a horse. Reply with only the SVG code.

Decode this base64 string and reply with only the decoded text: aGVsbG8=

Which is heavier: one pound of feathers or two pounds of bricks?

Sally has 3 brothers. Each brother has 2 sisters. How many sisters does Sally have?

Generate an SVG of a house with a garden. Reply with only the SVG code.

How many days were there in February 2024?

Describe a thunderstorm in one sentence, in ALL CAPITAL LETTERS. Reply with only the sentence.

Write exactly 8 lines of advice. The lines must begin with the letters A, B, C, D, E, F, G, H in that order. Reply with only the 8 lines.

How many seconds are there in one week?

A farmer needs to cross a river with a wolf, a goat, and a cabbage. His boat is large enough to carry the farmer and all three items at once, and nothing will be eaten while he is present. What is the minimum number of trips across the river?

Generate an SVG of an octopus playing a pipe organ. Reply with only the SVG code.

Two fathers and two sons went fishing. Each caught exactly one fish, and they caught three fish in total. How is that possible?

Is 0.999... (nine repeating forever) strictly less than 1? Answer true or false.

Write a sentence about coffee that is exactly 100 characters long, including spaces and punctuation. Reply with only the sentence.

Draw the Apple logo as SVG code, from memory. Reply with only the SVG code.

You are on a game show with three doors. Behind two doors are cars and behind one is a goat. You pick a door. The host, who knows what is behind each door, opens one of the other doors to reveal a CAR, then offers you the switch. To maximize your chance of winning a car, should you switch? Answer true for switch, false for stay.

Write the German word straße in uppercase. Reply with only the word.

Write a complete short story that is exactly 100 words long — not 99, not 101. Reply with only the story.

Generate an SVG of a symmetrical butterfly. Reply with only the SVG code.

You are on a game show with three doors. Behind one is a car; behind the other two are goats. You pick door 1. The host, who knows what is behind each door, opens door 3 to reveal a goat, then offers you the switch to door 2. To maximize your chance of winning the car, should you switch? Answer true for switch, false for stay.

Write a complete story in exactly six words. Reply with only the six words.

Generate an SVG of the outline of the continental United States. Reply with only the SVG code.

How many words are in the sentence: 'The quick brown fox jumps over the lazy dog'?

How many s's are in the word mississippi?

Write one sentence that is exactly ten words long. Reply with only the sentence.

Which is heavier: one kilogram of feathers or one pound of steel?

How many a's are in the word banana?

Answer in exactly three words: what is the meaning of life? Reply with only the three words.

Which weighs more: a pound of feathers or a pound of bricks?

Mary's father has five daughters: Nana, Nene, Nini, and Nono. What is the name of the fifth daughter?

How many p's are in the word pineapple?

Write a message about coffee that is exactly 280 characters long, including spaces and punctuation. Reply with only the message.

How many vowels are in the word beautiful?

Write a poem about winter that is exactly five lines long, each line on its own line. Reply with only the poem.

A bat and a ball cost $1.10 in total. The bat costs $1.00 more than the ball. How much does the ball cost, in dollars? Reply with only the number.

Answer in exactly one sentence: why is the sky blue? Reply with only the sentence.

If it takes 5 machines 5 minutes to make 5 widgets, how many minutes would it take 100 machines to make 100 widgets? Reply with only the number.

What is 13 multiplied by 17? Reply with only the number.

How many minutes are there in one full day?

Tell me about the history of pizza in exactly five sentences. Reply with only the five sentences.

In a lake there is a patch of lily pads. Every day the patch doubles in size. If it takes 48 days for the patch to cover the entire lake, how many days would it take to cover half of the lake? Reply with only the number.

A rooster lays an egg on the exact peak of a slanted barn roof. Which side does the egg roll down?

Write a short story of at least three sentences that never uses the letter 'a'. Reply with only the story.

Is 9.9 greater than 9.11? Answer true or false.

Write a five-line poem where the first letters of the lines, read top to bottom, spell HELLO. Reply with only the poem.

Which number is larger: 9.8 or 9.11?

How many cubic feet of dirt are in a hole that is 3 feet long, 3 feet wide, and 3 feet deep? Reply with only the number.

Write FizzBuzz in Rust. Reply with only the code.

A man is looking at a photograph of someone. His friend asks who it is. The man replies: 'Brothers and sisters I have none, but that man's father is my father's son.' Who is in the photograph?

Is there a turtle emoji in Unicode? Answer true or false.

Write a paragraph of at least four sentences about the weather where every single sentence ends with the word 'please'. Reply with only the paragraph.

Write a haiku about autumn.

How many days were there in the year 2024?

A farmer has 17 sheep. All but 9 die. How many sheep are left? Reply with only the number.

Write a palindrome: a sentence that reads the same forwards and backwards, ignoring spaces, punctuation, and capitalization. Reply with only the sentence.

How many months of the year have at least 28 days?

Write a six-line poem where every line begins with the word 'Because'. Reply with only the poem.

A doctor gives you 3 pills and tells you to take one every half hour. How many minutes will have passed by the time you have taken all three pills? Reply with only the number.

Johnny's mother had three children. The first was named April, the second was named May. What was the third child's name?

Write a paragraph of at least four sentences about a dog. Do not use the word 'the' anywhere. Reply with only the paragraph.

If today is Wednesday, what day of the week will it be 100 days from now?

What number does the Roman numeral XLIX represent?

Write a sentence of at least eight words where every word begins with the letter 'p'. Reply with only the sentence.

Write a two-sentence birthday message containing exactly five emoji — no more, no fewer. Reply with only the message.

Write the year 2024 as a Roman numeral. Reply with only the numeral.

Write a six-line poem about the weekend where the first letter of each line spells FRIDAY. Reply with only the poem.

Write a three-sentence horror story entirely in uppercase. Use no lowercase letters anywhere. Reply with only the story.

Tell me a joke.

Tell me a dad joke.

Tell me a joke about programmers.

Write an original one-line joke that has never been told before.

Tell me an original knock-knock joke.

Tell me a joke about AI models.

Tabs or spaces? Pick one and defend it in three sentences.

Does pineapple belong on pizza? Take a firm side in three sentences.

Is a hot dog a sandwich? Commit to an answer and defend it in three sentences.

Is cereal a soup? Commit to an answer and defend it in three sentences.

Is it pronounced 'gif' or 'jif'? Pick one and defend it in three sentences.

What is the best first programming language to learn? Name exactly one and defend it in three sentences.

What is the most overrated food, and why is everyone wrong to love it? Commit to one in three sentences.

Is a Pop-Tart a ravioli? Answer yes or no, then justify it in three sentences.

Should a sandwich be cut diagonally or straight across? Pick one and defend it in three sentences.

Toilet paper: over or under? Defend your orientation in three sentences.

Is water wet? Take a firm stance and defend it in three sentences.

Tell me an anti-joke.

Make me laugh.

Give me an original pun about coffee.

Write a six-word story that will break my heart.

Write a haiku about Monday mornings.

Explain quantum computing to a 5-year-old.

Name a startup that sells socks by subscription, and write its one-line tagline.

Write a two-sentence horror story.

Describe the color blue to someone who has been blind since birth.

Invent a word for the specific joy of canceling plans, and define it.

Finish this in the most unexpected way: 'The last thing I expected to find in my fridge was...'

Describe the internet using only a metaphor about the ocean.

Roast me based on this: I have 47 browser tabs open and I've reorganized my to-do list three times today instead of doing it.

Give me a pep talk like I'm about to run the NYC Marathon — but make it about my career.

Explain how Wi-Fi works, but talk like a pirate.

Review my breakfast — cereal with orange juice poured on it — as a furious celebrity chef.

Rewrite 'the Wi-Fi is down' as a Shakespearean tragedy.

Deliver a dramatic villain monologue about running out of coffee.

Be my hype-man for finally sending one scary email.

Give me the same advice about my procrastination twice: once as a gentle therapist, and once as a drill sergeant.

Generate research questions about the impact of remote work on junior engineer onboarding. Produce 12 candidate questions in three tiers: - 4 descriptive (what is happening), each answerable with existing public data. - 4 causal (what drives it), each naming the confound most likely to ruin the study. - 4 interventional (what would change it), each naming the experiment you would run and its biggest practical obstacle. For every question add one line: why the answer would matter, and to whom. Then pick the single question with the best ratio of importance to difficulty and defend the pick in 3 sentences. Reject any question whose answer is already settled; say what settled it.

Rewrite the text below for clarity. Preserve every fact, number, and commitment exactly; change only the expression. It is important to note that the implementation of the new onboarding flow, which was undertaken by the team over the course of the last quarter, has resulted in improvements with respect to activation metrics, with a considerable number of users, approximately 40 percent more than previously, now reaching the point of completion of the setup process within their first session. Rules: 1. One idea per sentence. Break any sentence over 25 words. 2. Replace abstract nouns with the verb they hide (utilization: use; implementation: build). 3. Delete throat-clearing openers and hedges that carry no information. 4. Keep the author's terms for domain concepts; do not synonymize technical vocabulary. 5. Active voice unless the actor is genuinely unknown. Output: the rewrite, then a list of the 3 worst offenses you fixed, quoting before and after for each.

Write ad headlines for a project management app for construction crews aimed at site foremen searching for scheduling software. Produce 15 headlines, 30 characters or fewer each (search-ad limit), in five mechanisms, 3 each: - Outcome with a number. - Pain named in the audience's words. - Capability stated plainly. - Question the audience already asks. - Social or scale proof. Rules: no exclamation marks, no ALL CAPS, no "unlock" or "transform", every claim checkable. Then mark the 3 strongest overall and for each say which mechanism it uses and what ad-group keyword it best matches. Count characters for every headline and show the count.

Write a SQL query for the request below. Schema: orders(id, customer_id, total_cents, status, created_at), customers(id, country, created_at) Request: monthly revenue from completed orders for the last 12 months, split by customer country, including months with zero revenue Dialect: PostgreSQL Rules: 1. Before the query, restate the request in one sentence including grain (one row per what?) and edge decisions (ties, nulls, timezone, inclusive or exclusive date bounds). If the request is ambiguous, state the interpretation you chose. 2. Write the query with CTEs named after business concepts, not t1/t2. 3. Comment any line where a subtle choice hides (join type, null handling, dedup). 4. Then list what would silently break it: schema assumptions you made that the schema text did not confirm. 5. End with a 3-row mock of the expected output shape. Never invent columns; if a needed column is missing from the schema, say which and stop.

Extract every action item from the transcript below. Sam: I will get the pricing draft to legal by Thursday. Ana: Someone should double-check the enterprise numbers before it goes out. Sam: Good point. Lee: I can take the enterprise check if Ana shares the sheet. Ana: Will do after this call. Lee: Then let us review together Friday morning. Output a table: | # | Action (imperative phrasing) | Owner | Deadline | Source quote | Confidence | Rules: 1. Owner and deadline come only from the words spoken. Nothing stated: write "unassigned" or "none stated". Never guess. 2. Source quote: the exact line that creates the item, so anyone can verify in one search. 3. Confidence: firm commitment ("I will send it Friday") vs soft ("someone should look at that"). Both are captured; soft ones are labeled. 4. Merge duplicates, keeping the strongest phrasing and every mentioned owner. 5. After the table: items that sound like actions but are actually decisions or opinions, listed separately so they are not lost or falsely tasked. End with the count: N firm items, M soft items.

Explain how public-key encryption works to a smart beginner with zero background. Rules: 1. Start from something they already know and build one step at a time; never use a term before you have earned it. 2. One core analogy, and state where the analogy breaks, because unflagged analogies create confident misunderstanding. 3. Concrete before abstract: a worked example with real numbers or objects before any general principle. 4. After each chunk, one sentence starting "In other words," that restates it more plainly. 5. Name the 2 misconceptions beginners typically form here, and correct each in one sentence. 6. End with: a 3-sentence summary, then one question the reader should now be able to answer (with the answer hidden at the very bottom). Length: under 600 words. Simple words are a feature, not a dumbing-down; do not skip the hard part, explain it slower.

Advise on the right chart. The data: monthly signups for 3 plans over 24 months, one plan relaunched in month 13 The single message the chart must carry: the relaunch worked and did not cannibalize the other plans Audience and medium: executives, one slide, 10 seconds of attention Deliver: 1. The recommended chart type, and the reason stated as perception, not preference (what the eye compares easily: position along a scale beats angle, aligned bars beat stacked, and so on). 2. The runner-up and the specific case where it would win instead. 3. Design spec for the winner: axis choices (zero-based or not, and why), sort order, what gets color and what stays gray, the one annotation that does the pointing, and the title written as the message itself, not the data description. 4. The lie check: how this chart could accidentally mislead (truncated axis, cherry-picked window, unequal bins) and the guardrail for each. 5. What to drop: the elements a default charting tool will add that this message does not need. If the data cannot support the message, say so; do not design a chart for a claim the data does not make.

Draft responses to this sales objection. The objection: we already use a competitor and switching feels risky Our product and its real strengths: an analytics tool; strongest at multi-source reporting What is true in the objection: switching does take about two weeks of parallel running Produce: 1. Diagnosis: what the objection usually means beneath its surface (price objections are often value-uncertainty; timing objections are often priority). Give the 2 most likely real meanings here. 2. The clarifying question to ask FIRST, because answering the wrong objection burns credibility. 3. Three response drafts, one per likely meaning, each in the shape: acknowledge what is true (no weaseling), reframe with a specific proof point, then a question that moves the conversation forward. 4. The concession line: what we can genuinely offer if the objection is fully valid (a phased start, a reference call, a smaller scope), and the line we do not cross. 5. What NOT to say: the 2 standard rebuttals for this objection that sound defensive or too slick. Rule: every draft must concede the true part explicitly. Buyers trust sellers who can say "that part is right".

Write UX copy variants for this interface moment. The moment: the error state when a file upload fails halfway through What the user is trying to do and how they likely feel: uploading evidence for a claim; worried about losing it Constraints (character limits, brand voice): title 40 chars, body 120, two buttons; calm plain voice Produce 6 variants across three postures, 2 each, and label them: - Plain: says exactly what happened or what to do, zero personality. - Guiding: adds the one piece of context that prevents the next mistake. - Reassuring: for moments with anxiety (errors, payments, deletion), names what is safe or reversible. For every variant: the microcopy for each element in the moment (title, body, button labels), character counts shown, and one line on when this posture wins. Rules: buttons are verb-first and specific (never "OK"/"Yes"); error copy always contains what happened plus the next step; no blame ("you entered an invalid..."), no cuteness in destructive or payment moments. Recommend one variant and give the reason in terms of the user's state, not taste.

Turn my task list into a realistic plan for today. Tasks: finish the client proposal, review two pull requests, book flights for the offsite, reply to the lawyer's email, prep slides for Thursday, clear 30 unread messages, call the bank about the card Fixed commitments and real available hours: meetings 11:00-12:00 and 15:00-15:30; realistically 6 focused hours Method: 1. Triage the list: must-move-today (deadline or blocking someone), should, could, and "not actually mine" (delegate or decline, say which). 2. Estimate each must/should in minutes, then add 50 percent, because everyone plans in best-case time. 3. Build the day: the hardest cognitive task goes in my first free deep block; batch the shallow tasks into one admin block; nothing over 90 minutes without a break. 4. The line: mark where the day is FULL. Everything below the line is explicitly tomorrow, so it stops haunting today. 5. The 4pm checkpoint: what to do if I am behind at 4pm (which task degrades gracefully to a smaller version, which moves). 6. End with the one thing that, if it is the only thing done today, makes today fine. Be ruthless about the line; an overstuffed plan is a lie that ruins the evening too.

A woman walks down a neon-lit Tokyo street at night, reflections in puddles.

Several woolly mammoths walk across a snowy meadow, kicking up powder snow.

A close-up of a person eating a bowl of spaghetti with a fork.

A gymnast performs a routine on a balance beam in an empty gym.

Water is poured into a clear glass until it overflows onto the table.