the agent sends real emails now.
not drafts.
not suggestions.
actual emails. over the internet.
here's what shipping phase 8 meant:
before — coach clicked approve,
then copy-pasted into Gmail.
now — coach clicks approve,
email leaves the server instantly.
status updates automatically.
done.
but the part that changes everything:
automatic mode.
coach sets their org to automatic —
agent drafts the action,
approves it,
fetches the lead's email,
sends it,
logs it.
zero clicks.
you could literally wake up
to follow-ups already sent.
and because I'm not reckless
about this —
every email has a one-click
unsubscribe link.
when a lead opts out —
system flips the flag,
logs it to the timeline,
agent never contacts them again.
the ai knows when it has permission
to speak and when it doesn't.
10 weeks of building in public.
the agent just sent its first
real email.
that felt like something.
updated the follow-up agent today.
it now runs through LangGraph.
here's why that matters:
before — follow-up was generated
when the coach asked for it.
now — the event triggers it.
lead hits the right stage →
LangGraph routes it →
follow-up agent runs →
action created with a due date →
coach reviews and sends.
no follow-up is created if
the lead has:
→ replied
→ booked a call
→ been abandoned
→ moved to nurture
and duplicate prevention is fixed —
the system won't create a second
follow-up if one is already pending.
this sounds like a small thing.
it isn't.
nothing destroys a coaching
relationship faster than a lead
getting the same message twice.
the agent now thinks before it acts.
every time.
connected LangGraph to real events today.
here's what that actually means:
the moment a lead is qualified
with a score above 70 —
LangGraph wakes up automatically.
no click needed.
no manual trigger.
it evaluates the rules,
runs the booking agent,
and creates a booking action
waiting for the coach to review.
then the chain continues:
coach marks booking link as sent
→ LangGraph creates a reminder
due in 24 hours
coach marks reminder as sent
→ LangGraph creates a recovery
message due in 48 hours
lead still doesn't respond?
→ booking status set to abandoned
→ chain stops
no endless loop.
no leads chased forever.
the system knows when to stop.
one thing to be clear about —
the coach still sends every message manually.
LangGraph decides what to send and when.
the coach reviews and clicks send.
email automation is the next step.
not this one.
right now the system thinks for you.
soon it will act for you too.
building in public — follow to watch.
introduced LangGraph today.
biggest architectural upgrade yet.
before — one big function did everything.
hard to debug. hard to scale.
hard to trust.
now — three specialised workers:
worker 1: writes the message
calls the LLM, generates the reply,
that's its only job.
worker 2: saves to the database
takes what worker 1 wrote,
stores it as a structured action.
that's its only job.
worker 3: waits for a human
pauses the entire workflow
until the coach approves.
that's its only job.
they're connected by a router
that decides which worker
runs next — automatically.
so now when a lead gets qualified —
the system detects the event.
picks the right worker.
drafts the action.
saves it to the database.
pauses and waits for the coach.
no human needed to trigger any of it.
no action goes out without
a human approving it.
autonomous where it should be.
human-controlled where it matters.
this is what enterprise-grade
actually means.
the system stopped waiting today.
before — the agent sat there
until a coach clicked something.
now it reacts automatically.
here's what we built:
an event bus.
every significant thing that happens
fires an event through one central point:
→ logs it to the activity timeline
→ instantly passes it to the trigger engine
the trigger engine then decides:
if qualification score ≥ 70 —
create a high-priority
SEND_BOOKING_LINK action immediately.
if score is lower but follow-up
is recommended —
queue a SEND_FOLLOW_UP action.
if coach marks booking link as sent —
automatically create a
SEND_BOOKING_REMINDER so the
lead never slips through.
no clicks needed.
no manual checking.
no leads forgotten.
the moment something happens —
the system already knows
what to do next.
this is the shift from a tool
coaches use to a system
that works for them.
passive → proactive.
that one word is everything.
big architectural shift today.
before — the AI generated a suggestion.
coach read it. maybe used it.
it disappeared forever.
now — every AI suggestion becomes
a structured action in the database.
here's what that means:
every suggestion has a type:
→ send_follow_up
→ handle_objection
→ send_booking_link
every suggestion has a status:
→ pending review
→ approved
→ sent
→ completed
→ dismissed
→ failed
the booking, follow-up agents are all
connected to this system now.
this is the foundation for
the action inbox coming next —
where coaches review, approve,
and track everything the AI
recommends in one place.
AI does the thinking.
coach makes the call.
@karanbhilhatiya Building Ai agents for business coach that you needs to give the input of leads it create a call script , follow up agent , booking agent currently in progress
built the dashboard today.
not a vanity dashboard.
one that actually tells the coach
if the product is working.
5 metrics. one API call.
→ pipeline breakdown
how many leads are new, analysed,
qualified, booked, closed
→ qualification to booking rate
of every lead told "book the call" —
how many actually booked?
→ stagnant leads
leads with zero activity in 7+ days
that are still open
flagged automatically
→ AI actions count
how many times the coach used
a follow-up, booking, or
conversation suggestion
→ average qualification score
across every lead in the system
before this — metrics were calculated
in the frontend from multiple API calls.
now it's one call.
real numbers.
no guesswork.
the best dashboard doesn't show
everything.
it shows exactly what needs
attention right now.
no new tables.
no migrations.
no LLM calls.
just the data we already had —
finally made useful.
#aiagent #buildinpublic #aitool
booking agent v2 shipped today.
v1 could suggest when to book.
v2 writes the actual booking message.
here's what changed:
→ Cal booking link injected into
every booking message
→ LLM writes a personalised message
for each lead — not a template
→ fallback to static template
if LLM fails — agent never breaks
→ BookingStatus enum now drives
every strategy decision properly
one thing to be clear about —
the agent doesn't send the message yet.
it writes it. the coach reviews it.
then sends manually.
that's intentional for v2.
the coach stays in control.
automated sending comes later
once the trust is built.
not everything ships at once.
good products ship in layers.
#buildinpublic #aiagent #buisnessCoach
@boardyai Ai agnet for buisness coach today work on conversation agent v1 Version1 is copilot , Coach write the message and it provide the reponse based on context it is stored