uma startup levantou $21m e passou o anúncio inteiro mostrando o produto.. um agente que cria seus anúncios, liga pros seus leads e agenda as reuniões aula de lançamento
We raised $21M to help small businesses beat the giants
We're giving $1M of it back to you today
Today we're launching Runable Grow: world's first agent that runs your entire go-to-market. 24/7
The last few months have been unreal.
Millions of users came to @runable_hq and built their entire business on it, websites, databases, payments, analytics, pitch decks, product videos & scheduled work.
Every single one of them asked the same question:
"How do I grow it?"
So we built the answer.
Runable Grow handles GTM end-to-end:
→ Runs ads on Meta, Google, and ChatGPT without any account, fully managed by Runable
→ Cold calls + cold emails from a real phone number and your business inbox
→ Generates and qualifies leads
→ Social listening
→ SEO + AEO optimization
Not a dashboard. Not a "copilot."
An agent that does the work.
It's a world-class media buyer, SDR, SEO lead, brand analyst working for your business while you sleep.
Don't believe me? Prove me wrong.
Your first $100 on Grow is on us. Every paid user. No catch.
Go break it 👇 https://t.co/92GPjhRyQA
Uma ferramenta de IA de US$ 20 por mês parece barata.
Dez ferramentas esquecidas custam US$ 2.400 por ano.
Uma vez por mês, eu abriria a fatura e perguntaria: qual etapa cada assinatura eliminou nas últimas quatro semanas?
Se a resposta for “ainda estamos explorando”, o teste já durou tempo demais.
O Gemini 3.8 Flash mantém o preço de US$ 0,75 por milhão de tokens de entrada e US$ 3,75 na saída. Mesmo assim, os primeiros relatos apontam um custo por tarefa cerca de 40% maior porque o modelo usa mais tokens e faz mais etapas de avaliação.
Esse detalhe importa mais do que a tabela de preços.
Empresas costumam comparar modelos pelo valor do token e pela nota no benchmark. Um agente cobra a conta inteira: chamadas de ferramentas, contexto repetido, tentativas, validações e tamanho da resposta. Dois modelos com o mesmo preço nominal podem terminar o mesmo trabalho com custos completamente diferentes.
Eu testaria o 3.8 em um conjunto fixo de tarefas reais e registraria custo por conclusão aprovada, tempo total e quantidade de correções humanas. Se ele gastar 40% a mais e reduzir pela metade as revisões, a troca pode valer muito. Se a qualidade melhorar apenas no benchmark, você comprou uma conta maior com o mesmo workflow.
O preço do token cabe numa landing page. O custo de operação só aparece quando o agente termina o trabalho.
@magronees O ganho real não está só em cortar mais rápido, mas em testar 10 ganchos a partir do mesmo vídeo e medir retenção. Você inclui no fluxo alguma etapa para escolher os trechos pelos dados, ou a seleção ainda depende do feeling?
@Its_Saad_kkk Centralizar quatro ferramentas só gera ROI se o contexto também atravessa o fluxo. Se eu ainda preciso copiar prompt, referência e briefing entre módulos, virou apenas um menu mais bonito.
O Seedance 2.5 agora está disponível no @itsPolloAI
Dessa vez testei os 30 segundos com algo um pouco mais caótico: um entregador tem sua pizza roubada por um macaco e decide correr atrás dele.
O mais impressionante é a consistência durante toda a sequência: o mesmo personagem, macaco, caixa de pizza e ambiente continuam coerentes enquanto a história fica cada vez mais absurda.
E o Seedance 2.5 está com 60% OFF por tempo limitado no Pollo AI.
Prompt:
Create a single coherent ultra-realistic 30-second comedic smartphone video in which an ordinary food delivery unexpectedly turns into a bizarre but believable chase after a monkey steals the delivery.
The entire video should feel like a spontaneous viral video captured on a modern smartphone by a random person standing nearby.
It must NOT feel like an advertisement, scripted comedy sketch, Hollywood movie, polished commercial, or generic AI video.
The humor comes from one increasingly absurd situation being treated completely seriously.
The sequence must tell ONE continuous story with a clear setup, escalation and final payoff.
MAIN CHARACTER:
A male food delivery rider around 25 years old.
Short dark hair, light beard, average athletic build.
He wears a plain black motorcycle helmet, black pants, dark sneakers, a distinctive BRIGHT ORANGE #FF4900 delivery jacket and a plain bright orange #FF4900 insulated delivery backpack.
Absolutely NO logos, company names, branding or readable text anywhere on his clothing or equipment.
The rider's face, body, clothing, helmet, jacket and backpack must remain consistent throughout the entire video.
SECOND CHARACTER:
One small realistic macaque monkey with natural brown-grey fur and realistic proportions.
It must behave like a REAL monkey rather than a cartoon character.
The same monkey must remain visually consistent throughout the entire sequence.
SETTING:
A warm residential neighborhood in a Mediterranean-looking city during late afternoon.
Narrow street, small houses and apartment buildings, balconies, parked scooters, trees, stone pavement and a few pedestrians in the distance.
Natural golden-hour sunlight.
SEQUENCE:
Start immediately in the middle of an ordinary delivery.
The delivery rider pulls up slowly on a small scooter and stops in front of a residential building.
He parks, opens the bright orange delivery backpack and takes out ONE closed cardboard pizza box.
As he turns toward the building entrance, a small macaque monkey suddenly drops naturally from a nearby low wall or tree branch.
The monkey notices the pizza box.
The rider notices the monkey.
For approximately one second, both stare at each other.
The monkey suddenly grabs the pizza box.
The monkey must physically grab THE SAME pizza box already being held by the rider. The box must NOT duplicate or teleport.
Show a believable brief struggle where both hold opposite sides of the same box.
The monkey wins the tug-of-war and escapes with the pizza.
The rider freezes for half a second, looks at his empty hands, then immediately starts chasing it.
The person filming begins following them with the smartphone.
The monkey runs down the sidewalk carrying the pizza box awkwardly but successfully.
The rider runs behind it.
The monkey jumps onto a low wall and runs along it while carrying the box.
It jumps down.
The rider nearly catches it.
The monkey quickly changes direction.
The rider almost slips while turning but naturally recovers.
Everything should remain just realistic enough that someone could initially believe this actually happened.
The monkey continues toward ANOTHER nearby residential building.
It reaches another house and climbs onto a low front gate while still carrying the pizza.
The exhausted rider catches up.
Then a completely unrelated man around 40 years old opens the front door.
The monkey calmly puts the pizza box down on the low wall directly in front of him and immediately climbs away onto a nearby tree or wall.
The homeowner looks at the pizza.
He looks at the monkey.
Then he looks at the exhausted delivery rider.
For the FINAL PAYOFF:
The confused homeowner slowly picks up the pizza box.
The exhausted rider looks at him.
They stare silently at each other for approximately one second.
Then the delivery rider, completely defeated, simply gives the homeowner a small thumbs-up.
The homeowner hesitates and gives him a confused thumbs-up back.
The monkey remains visible casually sitting on a wall or tree branch in the background.
Hold this absurd composition for approximately two seconds.
Then end the smartphone recording naturally.
CAMERA:
The entire sequence must feel captured by ONE person using ONE smartphone.
Use realistic handheld smartphone movement, natural autofocus, occasional focus hunting, automatic exposure changes, natural motion blur, minor rolling shutter, imperfect framing, subtle digital stabilization and realistic running-camera shake.
NO drone shots.
NO impossible camera positions.
NO cinematic crane shots.
NO camera teleportation.
NO slow motion.
NO perfectly stabilized gimbal footage.
COMEDIC DIRECTION:
This is DEADPAN comedy.
The video itself never acknowledges that it is supposed to be funny.
No exaggerated reactions, screaming, cartoon behavior, meme faces, subtitles, captions, emojis, generated text or comedic sound effects.
The monkey NEVER deliberately performs comedy for the camera.
Its natural behavior creates the comedy.
PHYSICAL REALISM:
Maintain realistic gravity, foot contact, acceleration, human body weight, monkey locomotion and hand-object interaction.
There is ONLY ONE pizza box.
The box remains rigid cardboard and maintains the same dimensions throughout.
The monkey's hands must visibly hold the box while carrying it.
The rider must show believable fatigue after running.
CONTINUITY IS EXTREMELY IMPORTANT:
Maintain EXACTLY the same delivery rider, facial features, hairstyle, helmet, #FF4900 orange jacket, black pants, shoes and delivery backpack throughout.
Maintain EXACTLY the same monkey throughout.
Maintain EXACTLY the same pizza box throughout.
Maintain the same neighborhood and coherent geography.
The chase must visibly connect the first location to the second.
No morphing. No duplicated characters. No duplicated monkeys. No duplicated pizza boxes. No changing clothes. No disappearing objects. No extra limbs. No floating. No teleportation. No random scene transitions. No fake readable text. No logos. No watermarks.
The final result should feel like ONE uninterrupted 30-second piece of smartphone footage documenting a completely ordinary food delivery that unexpectedly escalates into a ridiculous monkey chase before resolving with the monkey apparently delivering the pizza to the wrong house.
O Seedance 2.5 agora está disponível no @itsPolloAI
Testei a geração de vídeos de 30 segundos com uma ideia simples: um cara perde o ônibus e decide que vai pegá-lo de qualquer jeito.
O resultado ficou absurdo. O modelo consegue manter o mesmo personagem, roupa, ônibus e sequência de ações enquanto uma situação completamente normal vai ficando cada vez mais caótica.
E o Seedance 2.5 está com 60% OFF por tempo limitado no Pollo AI.
Prompt:
Create a single continuous ultra-realistic 30-second comedic video following the same young man desperately trying to catch a city bus after missing it by only a few seconds.
The video should feel like a spontaneous viral clip filmed by one of his friends on a modern smartphone — NOT like an advertisement, movie trailer, polished commercial, or staged comedy sketch.
The humor must come from a completely normal everyday situation escalating progressively into ridiculous physical commitment, while everyone else in the city behaves as if nothing unusual is happening.
MAIN CHARACTER:
A man around 25 years old, athletic but otherwise completely ordinary-looking, short messy dark brown hair, light stubble, wearing a distinctive plain BRIGHT ORANGE #FF4900 lightweight jacket with no logos or text, black T-shirt underneath, loose dark grey pants and white everyday sneakers.
His face, body proportions, hairstyle, orange jacket, pants and shoes must remain EXACTLY CONSISTENT for the entire 30-second video.
The orange jacket should naturally stand out against the environment and remain clearly recognizable throughout the sequence.
SETTING:
A realistic busy European city street during late afternoon.
Natural architecture, parked cars, pedestrians, bicycles, traffic lights, road markings, bus stops, storefronts and normal urban activity.
The bus is a completely ordinary modern city bus with NO visible company logo, advertisements or readable text.
The environment must remain geographically coherent. The character and bus are moving forward through the SAME neighborhood rather than teleporting between unrelated locations.
SEQUENCE:
Begin immediately with the funny situation. No establishing shot.
The smartphone camera is already recording from the sidewalk.
The young man runs toward a bus stop just as the bus doors close directly in front of him.
He reaches the door approximately one second too late.
He quickly taps the closed glass door with his hand and looks directly at the driver with a desperate "seriously?" expression.
The driver does not react.
The bus casually begins driving away.
The man freezes for approximately half a second.
He looks at the departing bus.
Then something changes in his expression.
Without saying anything, he decides:
"I'm catching that fucking bus."
He suddenly starts sprinting after it.
The friend filming immediately starts running behind him while continuing to record on the smartphone.
The footage becomes naturally shakier from running.
The man initially runs normally behind the bus.
The bus is moving slowly because of city traffic, making it JUST plausible enough that he can keep chasing it.
He narrowly avoids a pedestrian.
He jumps over a small street obstacle without stopping.
He cuts across the corner of the sidewalk.
His bright orange jacket moves realistically in the wind.
The situation should progressively become more ridiculous, but his facial expression remains DEAD SERIOUS.
He is treating catching this bus like an Olympic final.
A cyclist passes nearby.
For approximately two seconds, the man grabs onto the back of the cyclist's bicycle while still running, allowing himself to accelerate.
The confused cyclist looks backward at him.
The man releases the bicycle and continues sprinting.
The bus slows slightly ahead because of traffic.
He gets closer.
The camera operator is struggling to keep up, breathing and running behind him.
The framing becomes imperfect and occasionally almost loses the man before finding him again.
The man reaches the rear side of the moving bus.
Instead of stopping, he notices a low exterior structural ledge near the rear of the bus.
He grabs onto it.
For a brief ridiculous moment, his feet continue running against the road while his hands hold the bus.
Then he pulls himself upward with believable physical effort.
IMPORTANT: his body movement must respect realistic gravity, weight and momentum.
No superhero jump.
No floating.
No teleportation.
He awkwardly climbs upward using realistic handholds and physical leverage.
The friend filming stops running and points the smartphone upward in complete disbelief.
The man somehow manages to climb onto the ROOF OF THE BUS.
He crawls forward slightly on the roof and finally sits upright.
His orange jacket is clearly visible.
The bus continues casually driving down the street as if nothing happened.
For the final comedic payoff:
The man sits calmly on top of the moving bus.
He catches his breath.
He fixes his hair.
Then he casually checks his wrist as though looking at a watch, visibly annoyed that the bus made him late.
Hold this absurd image for the final 2–3 seconds.
The camera remains on the sidewalk watching the bus drive farther away with the tiny orange-jacketed man sitting on top.
The recording ends naturally.
CAMERA AND REALISM:
The entire sequence should feel captured spontaneously on a smartphone by the SAME friend following the action.
Use realistic smartphone characteristics: natural autofocus, occasional focus hunting, automatic exposure adjustments, rolling-shutter imperfections, realistic motion blur, subtle digital stabilization, handheld shake, imperfect framing, small accidental camera tilts and realistic running-camera movement.
Do NOT use impossible cinematic camera positions, drone shots, slow motion or dramatic Hollywood camera movements.
The person filming must physically remain somewhere plausible relative to the character.
PHYSICS:
Everything must have convincing real-world physics.
Running must show realistic acceleration, momentum and foot contact. Clothing must react naturally to movement and wind. Hands must correctly interact with objects. The bicycle interaction must have believable momentum. The bus must maintain believable speed and road contact.
The climb onto the bus must look difficult and physically awkward rather than effortless.
COMEDIC STYLE:
The comedy must remain completely DEADPAN.
Nobody laughs at the camera. No meme text. No subtitles. No exaggerated cartoon reactions. No visual punchlines added artificially. No comedic sound effects.
The joke is simply that an ordinary man misses his bus and responds with an increasingly insane amount of effort while behaving as though this is perfectly reasonable.
CONTINUITY IS EXTREMELY IMPORTANT:
Keep EXACTLY the same man, facial features, hairstyle, bright orange #FF4900 jacket, black shirt, grey pants and white sneakers throughout.
Keep exactly the same bus throughout.
Maintain coherent street geography.
No unexplained cuts. No character duplication. No clothing changes. No disappearing objects. No morphing faces. No extra limbs or fingers. No broken anatomy. No floating. No teleportation. No gibberish text. No logos.
The final result should feel like ONE coherent 30-second viral smartphone video in which a mundane event gradually becomes completely absurd.
Quanto mais autonomia você dá para um agente, maior fica o custo potencial de um único erro.
Uma resposta errada em um chatbot é um problema.
Uma resposta errada executada 10.000 vezes por dia é um problema de negócio.
Guardrails começam a fazer muito mais sentido quando você olha por esse lado.
Today we're launching Guard for Agents: the enforcement runtime for AI 🔥
Your agent: "Full refund approved, plus $500 for the trouble!"
No such policy exists. You're on the hook. Now multiply by 10,000 conversations a day.
Guard catches it, names the policy, and ships this instead:
"Your order is past our 30-day refund window, but I can offer store credit or a replacement."
Milliseconds, before delivery.
Ninguém quer virar especialista em aspirador só para comprar um aspirador.
Mesmo assim, hoje a gente abre dezenas de abas, assiste reviews, compara modelos quase idênticos e ainda termina com dúvida.
A IA pode transformar essa preguiça num mercado gigantesco. Você informa orçamento, tamanho da casa, animais e o que te incomoda. Ela elimina quase todas as opções e entrega duas.
Muita gente vai preferir uma escolha boa em dois minutos à escolha perfeita depois de uma hora.
@marvyndias Esse uso é muito melhor do que pedir uma resposta pronta. A IA vira um teste de estresse: encontra contra-argumentos, lacunas e premissas frágeis. O profissional continua decidindo, mas revisa com mais cobertura e menos tempo.
@kipperdev O nome muda, mas a conta continua a mesma: quanto trabalho humano existe por cliente e quanto disso pode ser repetido com margem. Se não escala sem contratar mais gente, ainda é serviço só ganhou um slide novo
Wan 3.0 no Magnific conseguiu transformar várias nuvens em pequenos personagens, e o mais engraçado é que cada uma parece ter uma personalidade diferente 😭
A ideia é super simples, mas funciona muito bem.
https://t.co/F4mh5DzMTr